**EXTENDED DEPTH OF FIELD USING A MULTI-FOCAL LENGTH LENS WITH A CONTROLLED RANGE OF SPHERICAL ABERRATION AND CENTRALLY OBSCURED APERTURE**

CHI WANLI (US)

GEORGE NICHOLAS (US)

CHI WANLI (US)

*;*

**G02B13/18***;*

**G02B3/02***;*

**G02B3/10***;*

**G02B9/00***;*

**G02B9/08***;*

**G02B13/00**

**G02B27/10**US20030057353A1 | 2003-03-27 |

CLAIM OR CLAIMS |

1 . An integrated computational imaging system for producing images having an extended depth of field comprising: a multifocal imaging subsystem for producing an intermediate image that is purposefully blurred; the multifocal imaging subsystem exhibiting spherical aberration as the dominant feature of the purposeful blur; a central obscuration that cooperates with the spherical aberration to reduce variance among blurring effects through a range of defocused positions; and a digital processing subsystem for deblurring the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image having an extended depth of field. |

2. The system of claim 1 in which an overall aberration of the multifocal imaging subsystem can be expressed as a phase delay function in nominal wavelengths of imaging light, and the spherical aberration contributes a phase delay of between 1 .8 and 6 wavelengths. |

3. The system of claim 2 in which the phase delay function can be expressed by a series in terms of (r/R) as follows: φ = a{r/Rf ± b {r/R)A ± c{r / Rf ± d{r I Rf + ... where φ is the overall phase delay, λ is a radius of a considered point within an aperture of the imaging subsystem, /? is a maximum radius of the aperture, and a, b, c, and c/are coefficients of the second, fourth, sixth, and eighth order terms respectively. |

4. The system of claim 3 in which the coefficient b of the fourth order term defines a third order spherical aberration. |

5. The system of claim 4 in which the coefficient b is greater than the coefficients c and d. |

6. The system of claim 4 in which the aperture of the multifocal imaging subsystem has a minimum radius δR, and the phase delays ø for r equal to δR and for requal to R are at least approximately equal. |

7. The system of claim 6 in which the overall aberration of the multifocal imaging subsystem is circularly symmetric. |

8. The system of claim 1 in which the multifocal imaging subsystem includes an annular aperture having a minimum radius δR corresponding to a radius of the central obscuration and a maximum radius R, and a ratio of δR/R is greater than or equal to 0.3. |

9. The system of claim 8 in which the ratio of δR/R is greater than or equal to 0.5. |

10. The system of claim 1 in which the multifocal imaging subsystem includes at least one lens that is designed substantially free of spherical aberration and a phase plate that produces the spherical aberration that forms the dominant feature of the purposeful blur. 1 1 . The system of claim 10 in which the phase plate is attached to an aperture of the multifocal imaging subsystem at an image plane side of the multifocal imaging subsystem. |

12. The system of claim 1 in which the multifocal imaging subsystem includes at least one lens element having a surface modified to produce the spherical aberration that forms the dominant feature of the purposeful blur. |

1 3. The system of claim 1 in which the multifocal imaging subsystem includes a plurality of imaging elements and the spherical aberration is divided among more than one of the imaging elements. |

14. An extended depth of field imaging system comprising: a multifocal imaging subsystem designed as a combination of an ideal imaging component and a spherical aberration component that balances in and out of focus effects through a range of object distances; an intermediate image detecting device for detecting images formed by the imaging subsystem exhibiting balanced in and out of focus effects through the range of object distances; and a computer processing device for calculating a recovered image having an extended depth of field based on a correction of the balanced in and out of focus effects through the range of object distances. |

1 5. The system of claim 14 in which the ideal imaging component provides for imaging an object at a given object distance and the range of object distances through which the spherical aberration component balances in and out of focus effects includes the given object distance. |

1 6. The system of claim 14 in which the ideal imaging component of the multifocal imaging subsystem includes a lens designed to contribute the ideal imaging component and a phase plate designed to contribute the spherical aberration component. |

1 7. The system of claim 16 in which the phase plate is located within an aperture of the multifocal imaging subsystem between the lens and the intermediate image detecting device. |

1 8. The system of claim 14 in which a beam of light for producing the intermediate image has a nominal wavelength, and the spherical aberration can be expressed as a phase delay in the amount of between 1 .8 and 6 wavelengths of the nominal wavelength of the beam of light. |

1 9. The system of claim 14 in which the phase delay can be expressed by a series in terms of (r/RJ as follows: φ = air I Rf ± b (rlRf ± c(r/R)6 ± d(r/R)λ + ... where φ is the overall phase delay, r is a radius of a considered point within an aperture of the imaging subsystem, R is a maximum radius of the aperture, and a, b, c, and tf are coefficients of the second, fourth, sixth, and eighth order terms respectively. 20. The system of claim 19 in which the coefficient b of the fourth order term defines a third order spherical aberration, and the coefficient b is greater than the coefficients c and d. |

21 . The system of claim 20 in which the multifocal imaging subsystem includes an annular aperture having a minimum radius δR and a maximum radius R in a ratio of δR/R equal to at least 0.3. |

22. The system of claim 21 in which the phase delays øfor requal to δR and for requal to R are at least approximately equal. |

23. A method of designing a multifocal imaging subsystem as a part of an integrated computational imaging subsystem comprising steps of: designing a first overall component of the multifocal imaging subsystem as an ideal imaging component for imaging an object at a given object distance; designing a second overall component of the multifocal imaging subsystem as a spherical aberrator for balancing in and out of focus effects through a range of object distances; combining the first and second components of the multifocal imaging subsystem to produce an intermediate image that is purposefully blurred and the second component of the multifocal imaging subsystem contributes an aspherical aberration that is the dominant feature of the purposeful blur; and arranging for supplying information concerning the intermediate image and the purposeful blur to a digital processing system for producing a recovered image having an extended depth of field. |

24. The method of claim 23 in which the range of object distances through which the spherical aberration component balances in and out of focus effects includes the given object distance. |

25. The method of claim 24 in which the range of object distances through which the spherical aberration component balances in and out of focus effects is at least six times larger than the object distances over which the first component produces diffraction-limited imaging. |

26. The method of claim 23 in which the step of designing the first component of the multifocal imaging subsystem includes designing an ideal lens for imaging the object at a given object. |

27. The method of claim 26 in which the step of designing the second component includes designing the second component to exhibit a phase delay between 1 .8 and 6 nominal wavelengths of imaging light. |

28. The method of claim 27 in which aberrations of the second component are defined by the phase delay function can be expressed by a series in terms of (r/R) as follows: φ = a{r/R)2 ± b {r/R)* ± c(r /R)6 ± d{r I Rf + ... where φ is the overall phase delay, /" is a radius of a considered point within an aperture of the imaging subsystem, /? is a maximum radius of the aperture, and a, b, c, and d are coefficients of the second, fourth, sixth, and eighth order terms respectively. |

29. The method of claim 23 including an additional step of arranging an aperture of the multifocal imaging subsystem with a central obscuration that is sized to cooperate with the second component of the multifocal imaging subsystem to further balance in and out of focus effects through a range of object distances. |

30. An integrated computational imaging system for producing images having an extended depth of field comprising: a multifocal imaging subsystem for producing an intermediate image of an object combining an ideal overall imaging component and a prescribed overall aberration component for purposefully blurring the intermediate image; a central obscuration of the multifocal imaging subsystem that renders point spread functions of object points produced with the purposeful blurring more uniform over a range of object distances; and a digital processing subsystem for deblurring the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image having an extended depth of field. |

31 . The system of claim 30 in which: (a) the central obscuration blocks a portion of an aperture of the multifocal imaging system having a radius R, (b) the central obscuration has a radius δR, and (c) the central obscuration is sized with respect to the aperture in a ratio of δR/R of 0.3 or more. |

32. The system of claim 31 in which the central obscuration is sized with respect to the aperture in a ratio of δR/R of 0.5 or more. |

33. The system of claim 30 in which each of the point spread functions of object points have a central peak and oscillating ring structures, and the central obscuration provides for narrowing an average point spread function among the object points. |

34. The system of claim 33 in which the central obscuration renders the oscillating ring structures more uniform among the object points. |

35. The system of claim 30 in which the ideal imaging component is provided by at least one conventional lens and the aberration component is provided by a phase plate. |

36. The system of claim 30 in which both the ideal imaging and aberration components are provided by a logarithmic asphere having a surface modified to produce the prescribed aberration. |

37. The system of claim 30 in which the prescribed aberration is a spherical aberration that is the dominant feature of the purposeful blur. 38. The system of claim 37 in which an overall aberration of the multifocal imaging subsystem can be expressed as a phase delay function in nominal wavelengths of imaging light, and the spherical aberration contributes a phase delay of between 1 .8 and 6 wavelengths. |

39. An integrated computational imaging system for producing images having an extended depth of field comprising: a multifocal imaging subsystem for producing an intermediate image that is purposefully blurred in such a way that renders point spread functions of object points more uniform over a range of object distances than corresponding point spread functions produced without the purposeful blur; an average of the point spread functions having a central peak; a central obscuration of the multifocal imaging subsystem providing for narrowing the central peak of the average point spread function among the object points; and a digital processing subsystem for deblurring the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image based on the average point spread function over a range of focal depths. |

40. The system of claim 39 in which the point spread functions have oscillating ring structures and the central obscuration renders the oscillating ring structures of point spread functions more uniform among the object points. |

41 . The system of claim 40 in which the central obscuration suppresses oscillating ring structures of the object points. |

42. The system of claim 39 in which: (a) the central obscuration blocks a portion of an aperture of the multifocal imaging system having a radius R, (b) the central obscuration has a radius δR, and (c) the central obscuration is sized with respect to the aperture in a ratio of δR/R of 0.3 or more. |

43. The system of claim 42 in which the central obscuration is sized with respect to the aperture in a ratio of δR/R of 0.5 or more. |

44. The system of claim 39 in which a multifocal imaging subsystem combines an ideal imaging component and a prescribed aberration component for purposefully blurring the intermediate image. |

45. The system of claim 44 in which the multifocal imaging system includes at least one conventional lens element that provides the ideal imaging component and a phase plate that contributes the prescribed aberration component. 46. The system of claim 44 in which the multifocal imaging system includes a logarithmic asphere that contributes both the ideal imaging and aberration components. |

47. The system of claim 39 in which a spherical aberration that is the dominant feature of the purposeful blur. |

48. The system of claim 47 in which an overall aberration of the multifocal imaging subsystem can be expressed as a phase delay function in nominal wavelengths of imaging light, and the spherical aberration contributes a phase delay of between 1 .8 and 6 wavelengths. |

49. The system of claim 39 in which the central obscuration includes a stop located between the multifocal imaging subsystem and the intermediate image. |

50. A multifocal imaging system for an extended depth of field comprising: a multiple lens element system that is purposefully blurred by means of a third order spherical aberration that contributes to an extended depth of field; and the third order spherical aberration being distributed among a plurality of the lens elements and forming a dominant monochromatic aberration of the system. 51 . The system of claim 50 further comprising a central obscuration that cooperates with the third order spherical aberration for reducing variation in the purposeful blur over the extended depth of field. |

52. The system of claim 51 in which the multiple lens element system includes an annular aperture having a minimum radius δR corresponding to a radius of the central obscuration and a maximum radius R, and a ratio of δR/R is greater than or equal to 0.3. |

53. The system of claim 50 in which the multiple lens element system includes at least two logarithmic aspheres. |

54. The system of claim 50 in which the third order spherical aberration can be expressed in phase delay in units of a wavelength of light intended for used with the multiple lens element system, and the third order spherical aberration produces a phase delay of between 1 .8 wavelengths and 6 wavelengths. |

55. An integrated computational imaging system for producing images having an extended depth of field comprising: an imaging subsystem for producing an intermediate image of an object; means for producing a purposeful blur in the intermediate image using a predetermined amount of third order spherical aberration that contributes to an extended depth of field; a digital processing subsystem for deblurring the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image having an extended depth of field. |

56. The system of claim 55 in which the third order spherical aberration balances in and out of focus effects through a range of object distances. |

57. The system of claim 55 in which an overall phase delay of the imaging subsystem can be expressed by a series in terms of (r/R) as follows: φ = a(r/Rf ±b (rlRf ± c(r / Rf ± d(r / Rf + ... where φ is the overall phase delay, r is a radius of a considered point within an aperture of the imaging subsystem, /? is a maximum radius of the aperture, and a, b, c, and dare coefficients of the second, fourth, sixth, and eighth order terms respectively. |

58. The system of claim 57 in which the coefficient b of the fourth order term defines the third order spherical aberration, and the coefficient /b is greater than the coefficients c and d. |

59. The system of claim 58 in which a beam of light for producing the intermediate image has a nominal wavelength, and the third order spherical aberration can be expressed as a phase delay in the amount of between 1 .8 and 6 wavelengths. 60. A method of designing a multifocal lens system for extended depth of field imaging comprising the steps of: formulating an imaging system as a combination of an ideal imaging component arranged for diffraction-limited imaging and an aberration component dominated by a third order spherical aberration; and determining an amount of the third order spherical aberration that reduces variations among impulse responses over the extended depth of field. |

61 . The method of claim 60 in which the step of determining the amount of the third order spherical aberration includes adjusting the amount of third order spherical aberration from one amount to another to identify more uniform combinations of impulse responses over the extended depth of field. |

62. The multifocal lens system made by the method of claim 60 in which the imaging system is purposefully blurred by mean of third order spherical aberration. |

63. A method of designing a multifocal lens system for extended depth of field imaging comprising the steps of modifying an ideal lens design by incorporating into the lens design an amount of third order spherical aberration; testing performance of the modified lens design for imaging over a range of focal depths; and adjusting the amount of the third order spherical aberration for producing point spread functions that vary less over the extended depth of field. |

64. The method of claim 63 in which the multifocal lens system includes a plurality of lens elements and the modification of the ideal lens design includes incorporating portions of the third order spherical aberration into more than one of the lens elements. |

65. The method of claim 63 including a further step of incorporating a central obscuration into the design to narrow the point spread functions. |

66. A method of designing an integrated computational imaging system for producing images having an extended depth of field comprising steps of: formulating an imaging system having an aberration component dominated by a third order spherical aberration for producing a blurred intermediate image; and determining an amount of the third order spherical aberration that reduces variations among impulse responses over the extended depth of field; calculating an impulse response that departs from an arithmetic average of the impulse responses over the extended depth of field to unevenly weight the impulse responses over the extended depth of field; and incorporating the calculated impulse response in a digital processing algorithm for deblurring the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image having an extended depth of field. |

67. The method of claim 66 including an additional step of incorporating a central obscuration in the imaging system for further influencing the impulse responses for producing a sharper recovered image. |

68. A method of recovering an image based on an intermediate image, comprising steps of: accessing an intermediate image; and performing an iterative digital deconvolution of the intermediate image using a maximum entropy algorithm including the sub-steps of: estimating a new image containing a combination of directional images; and altering the directional images using a metric parameter to speed convergence toward a recovered image while avoiding points of stagnation. |

69. The method of claim 68 in which the metric parameter has a non-integer value. |

70. The method of claim 69 in which the metric parameter has a value between 0.2 and 0.6. |

71 . The method of claim 68 in which the directional images include pixel values, and the metric parameter adjusts weights among the pixel values. |

72. The method of claim 71 in which the metric parameter is adjusted above zero to give larger pixel values more weight. |

73. The method of claim 68 in which the metric parameter has a value that results in a modulation transfer function having a shape that increases contrast at high spatial frequencies approaching a Nyquist limit. |

74. The method of claim 68 including a further step of producing the intermediate image using a multifocal imaging system. |

75. The method of claim 74 in which the intermediate image produced by the multifocal imaging system exhibits a point spread function having side lobe rings, and the metric parameter is adjustable within a range that reduces side lobe rings of the point spread function in the reconstructed image. |

76. The method of claim 74 in which the multifocal imaging system includes third-order spherical aberration. |

77. The method of claim 74 in which the multifocal imaging system includes a central obscuration. |

78. An iterative digital deconvolution method for image processing comprising steps of: calculating a succession of estimated images of an object based on a starting image and combination of directional images; inputting a metric parameter that modifies the directional images; evaluating the estimated images modified according to the metric parameter against a criterion of entropy maximization and a statistical noise parameter to choose a closer estimate of the object; and outputting an enhanced image of the object over the starting image. |

79. The method of claim 78 in which the estimated images contain pixel values and the metric parameter alters weights assigned to the pixel values. 80. The method of claim 79 in which the metric parameter has a non-integer value. |

81 . The method of claim 80 in which the metric parameter has a value between 0.2 and 0.6. |

82. The method of claim 78 in which the starting image contains a purposeful blur. |

83. The method of claim 82 in which spherical aberration is the dominant feature of the purposeful blur. |

84. The method of claim 83 in which the spherical aberration balances in an out of focus effects through a range of object distances. |

85. The method of claim 84 in which the enhanced image has an extended depth of field. |

TECHNICAL FIELD

[0001 ] The invention relates to imaging scenes or other objects that can be brought to focus by combining a blurred intermediate image with digital processing that produces a recovered image having an extended depth of field. Particular applicability is found for the invention in photographic applications, although other applications including pattern recognition, detection, microscopy, machine vision, and optical measurement can also benefit from the invention.

BACKGROUND ART

[0002] Objects imaged by conventional imaging subsystems are sharply in focus over a limited distance known as the depth of field, which is inversely proportional to the square of the imaging system's numerical aperture for diffraction-limited imaging. Present-day cameras have mechanical focusing means, including automatic systems, to provide high quality images of particular scenes at various object distances. Even with these means, it is difficult to photograph objects clearly that span a large range of such distances. Cameras with a larger depth of focus will clearly provide superior photographs.

[0003] Digital processing of image data on a pixel-by-pixel basis has afforded more opportunity for improving and correcting optically imaged scenes. ^Some of these improvements have related to increasing the depth of field. For example, digital processing has been used to combine images of the same scene taken at different depths of focus to produce a composite image having an extended depth of field. The multiple images take time to collect, are difficult to process, and are generally unsatisfactory for scenes subject to change.

[0004] Amplitude attenuation filters have also been used to extend the depth of field. Typically, the attenuation filters are located in the aperture of the imaging systems, leaving inner radii clear but attenuating the outer annulus. However, the filter introduces large amount of light loss, which limits its applications.

[0005] More promising attempts have been made that deliberately blur an intermediate image in a systematic way so that at least some information about the imaged object is retained through a range of focus positions and a non-ideal impulse response function remains substantially invariant over the defocus range. Digital processing, which effectively deconvolutes the point spread function, restores the image to a more recognizable likeness of the object through an extended depth of field.

[0006] One such example locates a cubic phase mask within the aperture of the imaging system to generate a distance invariant transfer function, thereafter. Digital processing removes the blur. Although significant improvement in the depth of field is achieved, the cubic phase mask is not rotationally symmetric and has proven to be expensive and difficult to fabricate.

[0007] Another such example similarly locates a circularly symmetric, logarithmic asphere lens to extend the depth-of-field, which is more economical to manufacture. However, for the log-asphere lens, the impulse response is not perfectly uniform over the full range of operation, and as a result, some degradation is experienced in the image quality of the recovered image.

[0008] Reconstruction algorithms for removing the blur of such intermediate images are subject to problems relating to the quality and efficiency of their results. Nonlinear processing algorithms can suffer from slow convergence or stagnation and produce images with reduced contrast at high spatial frequencies.

DISCLOSURE OF INVENTION

[0009] Our invention, in one or more preferred embodiments, incorporates a multifocal imaging subsystem that purposefully blurs intermediate images of objects such as by introducing a controlled amount of spherical aberration to provide a more uniform impulse response over a range of focal depths. Third order spherical aberration is preferably the dominant feature of the purposeful blur. A digital processing subsystem recovers images convoluted through the more uniform impulse response for producing likenesses of the objects that remain in focus through an extended range of object depths. The multifocal imaging subsystem that purposefully blurs intermediate images is preferably circularly symmetric with a centrally obscured aperture, which narrows the impulse responses and avoids conditions such as contrast inversion for larger amounts of defocus. Together, the controlled spherical aberration and the centrally obscured aperture provide a sufficiently narrow and invariant impulse response over the extended depth of focus for achieving diffraction- limited performance over the extended range far exceeding classical limits. Diffraction-limited performance over a depth of field increased by six to ten times the classical limit can be achieved. The circularly symmetric structure of the imaging subsystem simplifies manufacture and reduces overall costs.

[0010] One example of an integrated computational imaging system in accordance with the invention for producing images having an extended depth of field includes a multifocal imaging subsystem for producing an intermediate image that is purposefully blurred. The multifocal imaging subsystem exhibits spherical aberration as the dominant feature of the purposeful blur. A central obscuration cooperates with the spherical aberration to reduce variance among the blurring effects through a range of defocused positions. A digital processing subsystem deblurs the intermediate image produced by the multifocal imaging subsystem and calculates a recovered image having an extended depth of field.

[001 1 ] The overall aberration of the multifocal imaging subsystem can be expressed as a phase delay function in nominal wavelengths of imaging light, and the spherical aberration preferably contributes a phase delay of between 1 .8 and 6 wavelengths. The controlled measure of spherical aberration is preferably mainly of the third order and is independent of the wavelength of illumination, focal length, and best focus object distance.

[0012] The aperture of the multifocal imaging subsystem is preferably circularly symmetric, having a minimum radius δR, defining an outer limit of a central obscuration and an inner limit of an annular aperture, and a maximum radius R defining an outer limit of the annular aperture. Preferably, a ratio of δR/R is greater than or equal to 0.3, provided that the light loss is not excessive.

[0013] At least one lens of the multifocal imaging subsystem can be designed substantially free of spherical aberration, and a phase plate can be designed to produce the spherical aberration that forms the dominant feature of the purposeful blur. Often, the phase plate can be attached to an aperture of the multifocal imaging subsystem at an image-plane side of the multifocal imaging subsystem. Alternatively, the multifocal imaging system can include at least one lens element having a surface modified to produce the spherical aberration that forms the dominant feature of the purposeful blur and that is in a proper range for reducing impulse response variance. The spherical aberration can also be divided among more than one lens element to increase design flexibility. The phase delays produced within the aperture at δR and R are preferably at least approximately equal for the center of the designated object range. [0014] Another example of the invention as an extended depth of field imaging system includes a multifocal imaging subsystem designed as a combination of an ideal imaging component and a spherical aberration component that balances in and out of focus effects through a range of object distances. An intermediate image-detecting device detects images formed by the imaging subsystem, exhibiting balanced in and out of focus effects through the range of object distances. A computer processing device calculates a recovered image having an extended depth of field based on a correction of the balanced in and out of focus effects through the range of object distances.

[001 5] The ideal imaging component preferably provides for imaging an object at a given object distance, and the range of object distances through which the spherical aberration component balances in and out of focus effects includes the given object distance. For example, the multifocal imaging subsystem can include a lens designed to contribute the ideal imaging component and a phase plate designed to contribute the spherical aberration component. The phase plate can be located within an aperture of the multifocal imaging subsystem between the lens and the intermediate image-detecting device.

[0016] The invention can also be practiced as a method of designing a multifocal imaging subsystem as a part of an integrated computational imaging subsystem. A first component of the multifocal imaging subsystem is designed as an ideal imaging component for imaging an object at a given object distance. A second component of the multifocal imaging subsystem is designed as a spherical aberrator for balancing in and out of focus effects through a range of object distances. Combining the first and second components of the multifocal imaging subsystem produces an intermediate image that is purposefully blurred. The second component of the multifocal imaging subsystem contributes an aspherical aberration that is the dominant feature of the purposeful blur. Information concerning the intermediate image and the purposeful blur is supplied to a digital processing system for producing a recovered image having an extended depth of field.

[001 7] The range of object distances through which the spherical aberration component balances in and out of focus effects preferably includes the given object distance. In addition, the range of object distances through which the spherical aberration component balances in and out of focus effects is at least six times larger than the object distances over which the first component produces diffraction-limited imaging. An aperture of the multifocal imaging subsystem is preferably arranged with a central obscuration that is sized to cooperate with the second component of the multifocal imaging subsystem to further balance in and out of focus effects through a range of object distances.

[001 8] Another example of an integrated computational imaging system in accordance with the invention includes a multifocal imaging subsystem for producing an intermediate image of an object combining an ideal imaging component and a prescribed aberration component for purposefully blurring the intermediate image. A central obscuration of the multifocal imaging subsystem renders point spread functions of object points produced with the purposeful blur more uniform over a range of object distances. A digital processing subsystem deblurs the intermediate image produced by the multifocal imaging subsystem and calculates a recovered image having an extended depth of field.

[0019] Each of the point spread functions of object points preferably has a central peak and oscillating ring structures, and the central obscuration provides for narrowing the average point spread function either for the close-in points or the distance object points, depending on the design. In particular, the central obscuration preferably renders both the widths of the central peaks and the oscillating ring structures more uniform among the object points.

[0020] The purposeful blur of the multifocal imaging subsystem is preferably the starting point for rendering the point spread functions of object points more uniform over a range of object distances. The central obscuration preferably removes variant components of the point spread functions produced by the purposeful blur for narrowing the central peak of the average point spread function among the object points, especially the object distance corresponding to the center section of the imaging subsystem. The digital processing subsystem preferably calculates the recovered image based on the average point spread function. Increases in performance associated with the central obscuration are believed mainly due to the similarities of the point spread functions over the design range of object distances rather than from any direct increase of the depth of field that might otherwise accompany the use of a central obscuration in an ideal imaging system. In particular, the associated improvements in the depth of field, particularly for close-in distances, are believed mainly due to both the narrower central peak of the average point spread function and the similar oscillating ring structures of the point spread functions over the designed object range. These two factors lead to point spread functions that vary less with object distance, so that the average point spread function used in the digital processing can provide a significantly improved output.

[0021 ] Yet another example of the invention as an integrated computational imaging system for producing images having an extended depth of field includes an imaging subsystem for producing an intermediate image of an object and means for producing a purposeful blur in the intermediate image using a predetermined amount of third order spherical aberration that contributes to an extended depth of field. A digital processing subsystem deblurs the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image having an extended depth of field.

[0022] An example of the invention as a multifocal imaging system includes a multiple lens element system that is purposefully blurred by means of a third order spherical aberration that contributes to an extended depth of field. The third order spherical aberration is distributed among a plurality of the lens elements and forms a dominant monochromatic aberration of the system. Preferably, a central obscuration cooperates with the third order spherical aberration for reducing variation in the purposeful blur over the extended depth of field. The plurality of lens elements can include at least two logarithmic aspheres.

[0023] A method of designing a multifocal lens system in accordance with the invention for extended depth of field imaging includes formulating an imaging system as a combination of an ideal imaging component arranged for diffraction-limited imaging and an aberration component dominated by a third order spherical aberration. The amount of the third order spherical aberration is determined so as to reduce variations among impulse responses over the extended depth of field. Preferably, the amount of the third order spherical aberration is determined by adjusting the amount of third order spherical aberration from one amount to another to identify more uniform combinations of impulse responses over the extended depth of field.

[0024] Another method of designing a multifocal lens system in accordance with the invention for extended depth of field imaging includes modifying an ideal lens design by incorporating into the lens design an amount of third order spherical aberration. The performance of the modified lens design is tested over a range of focus positions. The amount of the third order spherical aberration is adjusted for producing point spread functions that vary less over the range of focus positions. A central obscuration is preferably incorporated into the design to narrow the point-spread functions, e.g., at the close-in distances for the β- design.

[0025] A method of designing an integrated computational imaging system in accordance with the invention for producing images having an extended depth of field includes formulating an imaging system having an aberration component dominated by a third order spherical aberration for producing a blurred intermediate image. The amount of the third order spherical aberration is determined so as to reduce variations among impulse responses over a range of focus positions. A calculated impulse response departs from an arithmetic average of the impulse responses over the range of focus positions to unevenly weight the impulse responses over the extended depth of field. The calculated impulse response is incorporated into a digital processing algorithm for deblurring the intermediate image produced by the multifocal imaging subsystem and for calculating a recovered image having an extended depth of field.

[0026] Thus, a circularly symmetric, spherically aberrated, multifocal imaging system with centrally obscured aperture can be used in accordance with the invention for intermediate imaging. The resulting impulse response is sufficiently invariant over a range of object depths to support digital processing for recovering an image that remains in focus over a range of focal depths. Without too much loss in light intensity, this combined system can produce diffraction-limited resolution over an extended depth of field that is ten times that obtained by a conventional lens system. Prior approaches to extending the depth of field have not had this capability. The new imaging system is also economical to manufacture, since it can be circularly symmetric.

[0027] The invention can also be practiced as a method of recovering an image based on an intermediate image, which includes accessing an intermediate image of a scene and performing an iterative digital deconvolution of the intermediate image using a maximum entropy algorithm. Using the maximum entropy algorithm, a new image is estimated containing a combination of directional images. These directional images are uniquely altered using a metric parameter to speed convergence toward a recovered image while avoiding points of stagnation.

[0028] The metric parameter reconciles conventional maximum entropy algorithms at metric parameter values of zero and one. Values of the metric parameter are preferably chosen between zero and one to advantageously adjust the weight of different pixel values. Preferably, the metric parameter has a value between 0.2 and 0.6. The appropriate choice of the metric parameter contributes to a modulation transfer function having a shape that increases contrast at high spatial frequencies approaching a Nyquist limit.

[0029] The intermediate image can be produced using a multifocal imaging system, such as an aspheric lens. Typical point spread functions of such lenses have oscillating bases, which reduce image contrast. The metric parameter is adjustable within its preferred range significantly reduces side lobe oscillation that is seen in the blurred image.

[0030] We refer to this new form of nonlinear digital processing as "metric parameter-maximum entropy algorithm" or the MPME algorithm, which is expected to have wide applicability to digital image processing. The attributes of rapid convergence and the avoidance of stagnation can be generally advantageous for image reconstruction, restoration, filtering, and picture processing.

BRIEF DESCRIPTION OF DRAWINGS

[0031 ] Fig. 1 is the block diagram of an integrated computational imaging system in accordance with the present invention.

[0032] Fig. 2 is the diagram of the multifocal imaging subsystem having a centrally obscured aperture.

[0033] Figs. 3A-3F are graphs depicting point spread functions of a centrally obscured /?-type multifocal lens for various amounts of spherical aberration.

[0034] Figs. 4A-4F are graphs depicting point spread functions of a non-centrally obscured /?-type multifocal lens for various amounts of spherical aberration.

[0035] Figs. 5A-5F are graphs depicting point spread functions of a centrally obscured γ-type multifocal lens for various amounts of spherical aberration. [0036] Figs. 6A-6F are graphs depicting point spread functions of a non-centrally obscured y-tγpe multifocal lens for various amounts of spherical aberration.

[0037] Fig. 7 is a top-level flow chart for nonlinear digital processing according to a maximum entropy algorithm.

[0038] Fig. 8 is a flow chart showing steps within the maximum entropy algorithm for determining successive estimations of the object imaged by the multifocal imaging system.

[0039] Fig. 9 is a graph plotting the curves that show a convergence advantage associated with an optimization of a metric parameter in the maximum entropy algorithm.

[0040] Fig. 10 is a set of images of two point objects separated by a diffraction-limited distance for an imaging subsystem having a full aperture, including: intermediate image (a) showing a diffraction-limited blurred image by an ideal lens for the point objects at the optimum object distance, intermediate images (b) (c) (d) showing blurred images by a spherically aberrated multifocal imaging subsystem for other object distances, and recovered images (e), (f), (g), (h) showing images recovered by the maximum entropy algorithm from the intermediate images (a) (b) (c) and (d), respectively.

[0041] Fig. 1 1 is another set of images of the two point objects separated by a diffraction-limited distance for an imaging subsystem having a centrally obscured aperture, including: images (a), (b), (c), (d), and (e) formed by an ideal lens with the central obscuration at different object distances, intermediate images (f), (g), (h), (i), and (j) formed by a spherically aberrated multifocal imaging system with the central obscuration at the same object distances, and recovered images (k), (I), (m), (n), and (o) showing images recovered by the maximum entropy algorithm from the intermediate images (f) (g), (h), (i), and (j), respectively.

[0042] Fig. 12 is a graph plotting recovered data for a two point object at a defocused object distance comparing results from a spherically aberrated imaging subsystem with different central obscuration values and the blurred image data for Nikon lens without a central obscuration.

[0043] Fig. 1 3 is a graph plotting recovered data for a two point object at an optimum object distance comparing results from a spherically aberrated imaging subsystem with different central obscuration values and the blurred image data for Nikon lens without central obscuration.

[0044] Fig. 14 is a set of images illustrating the maximum entropy recovery of a defocused tiger image, including image (a) formed by ideal lens without central obscuration and recovered images (b), (c), and (d) from a spherically aberrated imaging system with different central obscuration values of 0.0 R, 0.3/?, and 0.5/?, respectively.

[0045] Fig. 1 5 is a graph plotting the overall transfer functions of the integrated imaging system with centrally obscured aperture for six object distances. [0046] Fig. 1 6 is a graph depicting the relatively small difference between the overall transfer functions of the integrated imaging system using point object and edge object.

BEST MODE(S) FOR CARRYING OUT THE INVENTION

[0047] Our studies of circularly symmetric multifocal lenses have revealed that a controlled amount of spherical aberration provides a desirable distance-invariant blur that leads to superior depth-of-field imaging. Our preferred multifocal lens for extending the depth of field can be based on any standard imaging arrangement modified to incorporate a third order spherical aberration as well as higher order spherical aberrations. Such standard imaging arrangements include Petzval lenses, Cooke lenses, and double Gauss lenses.

[0048] In addition, our studies have found that a further improvement in depth of field imaging, particularly at diffraction-limited resolution throughout the extended depth of field, can be realized by obscuring a center portion of the aperture of the multifocal imaging subsystem to narrow the impulse response for close-in distances (β-design) or for far distances (y- design). This increases the range in distance over which the impulse response is invariant. The center portion of the multifocal imaging subsystem can be a major contributor to the variation in the impulse response with distance. The combination of a central obscuration with a properly designed multifocal imaging can be used to further extend the depth of field, or to support higher resolution imaging through the extended depth of field.

[0049] Referring to Fig. 1 , an integrated computational imaging system 1 0 for extended depth of field imaging includes a multifocal imaging subsystem 12, an intermediate image detection device 14, a digital processing subsystem 16, and a display 1 8.

[0050] The multifocal imaging subsystem 12 includes a single or multiple element lens 22 and a phase plate 24. The lens 22 is preferably a conventional lens having at least one spherical surface arranged for ideal imaging and the phase plate 24 is preferably arranged to contribute a predetermined amount of spherical aberration. A central obscuration 26 can also be located within an aperture 28 of the multifocal imaging subsystem 1 2 for further improving performance. The phase plate 24 can be fabricated separately and aligned with the lens 22 as shown, or the optical contribution of the phase plate can be incorporated into a surface of lens 22, such as in the form of a logarithmic lens. Although both the lens 22 and the phase plate 24 are preferably transmissive, either or both of the lens 22 and the phase plate 24 can be alternatively fashioned as reflective surfaces such as in telescopic photography applications. The central obscuration 26 can also be realized in different ways, such as by adding a central stop within the aperture 28 or by arranging for an annular pattern of illumination that has center darkness. A pre-existing central stop can also be used for purposes of the invention, such as a secondary mirror of a telescope. [0051 ] Other imaging systems contemplated by the invention include multiple lens elements such as for dealing with chromatic aberrations or other imaging requirements. The invention provides increased flexibility within such multiple lens element designs to distribute the desired amount of spherical aberration among a plurality of the lens elements. For example, at least two of the lens elements can be formed as logarithmic aspheres, each incorporating a portion of the desired spherical aberration.

[0052] The image detection device 14, which collects an intermediate image 30 of objects 20 that is generally blurred, can be fashioned as a pixilated CCD (charge coupled device) or CMOS (complementary metal oxide semiconductor) detector or other light sensitive device. The detector pixels can be arranged as a two-dimensional array, as a one- dimensional array, or even as a single detector pixel. Any pixel combinations short of two dimensions are preferably subject to scanning for collecting enough information to complete a two dimensional intermediate image 30. However, one-dimensional imaging can be used for particular applications.

[0053] The digital processing subsystem 16 preferably includes a computer-processing device having a combination of hardware and software for the purpose of image processing. The digital processing subsystem 1 6 can be incorporated into a camera system that also includes the multifocal imaging subsystem 1 2, or the digital processing subsystem 1 6 can be arranged as a standalone image-processing computer. The primary purpose of the digital processing subsystem 1 6 is to sharpen the intermediate image 30. An inverse filter or its modifications, e.g., Wiener filter, can be used for this purpose. Preferably, a nonlinear algorithm, such as an iterative maximum entropy algorithm, is used to sharpen the intermediate image 30. If maximum entropy algorithm is used, an optional acceleration factor, metric parameter, can be chosen to optimize the speed and convergence.

[0054] The digitally processed image, which is referred to as a recovered image 32, is outputted to the display device 1 8, which can be CRT (cathode ray tube), LCD (liquid crystal display) or other display device appropriate for viewing purposes. Alternatively, the display device 1 8 can be omitted and the recovered image 32 can be inputted to other functional hardware/software. For example, the recovered image 32 can be input to a pattern recognition system or a machine vision system. If the recovered image 32 is used for these latter purposes, then the digital processing subsystem can be incorporated into the pattern recognition or machine vision system. The digital processing device can become optional depending on the amount of blurs in the intermediate image.

[0055] The integrated computational imaging system 10 is applicable to binary or gray scale or color imaging. It is also applicable to a range of different wavelengths including infrared imaging.

[0056] An optical diagram of a modified multifocal imaging subsystem 1 2 for producing the intermediate image 30 is illustrated in FIG. 2 based on use of a logarithmic asphere 34 that combines the ideal imaging of the lens 22 with a predetermined spherical aberration of the phase plate 24. A point source 5, which is located along an optical axis 36 at a distance so away from the logarithmic asphere 34 at object plane I, is imaged as a blurred intermediate point image Pat image plane Il at a distance f along the optical axis 36 on the other side of the logarithmic asphere 34. The logarithmic asphere 34 is mounted within the annular (or ring-type) aperture 28, having a radius from δR to R, where δR is the radius of the central obscuration 26 and R is the radius of lens aperture 28, where 0 < δR < R. The center portion of the lens aperture 28 from the optical axis 36 to δR is blocked by the central obscuration 26 in the form of a disk- shaped stop. However, <5/?=0 is treated as a special case of a full aperture, which is consistent with a particular embodiment of the invention.

[0057] In optical system design, the ideal imaging components and the aberrations can be described as follows: φ/k = W + φldeal /k (D wherein the phase delay φ is measured in radians, Ar equals 2π/λo where λo is the average wavelength of illumination, and W\s the optical path difference (O.P.D.) in micrometers.

[0058] For the ideal imaging system, it is well known that φukai = k(ylt2 + r2 - 1 + φo2 + r2 - S0) (2) in which the phase delay φideai'\s measured in radians for a perfect diffraction limited lens, /-is the radial coordinate in plane I, and søis the focused object position, and k=2π/λo.

[0059] For the ideal lens, as an example with so=λ 500mm, f=62.5mm, /?=8mm, and λø=0.5μm, from a power series expansion of φideai, one can readily find: φideαl = 6702.07(A2 -26.35(A4 +0.21(A6 +... K K K (3) which Equation (3) is valid in the nonparaxial regime.

[0060] For the O.P.D. W, of the two types of logarithmic aspheres denoted by subscript β and subscript y, one can express them as follows:

W = J s2+ r2 -s + — ώ lπ (4) where so is the center of the depth of field range, λo is the wavelength of illumination in vacuum, and the expression of φpfor the two types of logarithmic aspheres, φpβ and φpy, can be written as follows. φPβ(p) = α/-~-{\n[Aβ(t2+r2)]-l}-α/j[]n(A/)-l) (5) where:

(6) and φPγ(p) = -aγ + r2)]-l} + ar -l] (7) where:

(8) [0061 ] From a power series expansion of Equations (5) or (7), it can be appreciated for purposes of the invention that additional spherical aberration is the dominant feature of the purposeful blur that is being introduced. This will be made more evident within a description of some specific embodiments.

[0062] For completing a design based on Equations (4)-(8), the desired range for the depth-of-field s/, Si- can be selected along with representative values for t, R, δR, so, and λo. Thereafter, the variables aβ, Aβ, and φpβ (or aγ, Aγ, and φpγ) can be computed. From these, Equation (4) can be used to compute the aberration term W.

[0063] The logarithmic aspheres described above are examples of a multifocal lens that can be constructed in accordance with the invention. From a more general point of view, a multifocal lens useful for extended depth-of-field imaging can be composed of any standard imaging arrangement that is designed to incorporate a predetermined amount of spherical aberration, including third order spherical aberration as well as higher order spherical aberrations. For example, such standard imaging and projection arrangements such as Petzval lenses, Cooke lenses, and double Gauss lenses can be used for these purposes.

[0064] For describing a multifocal lens in terms of a range of aberrations, it is useful to expand the aberration function a series in terms of (r/R). For example, if the design parameters: sø= l 500mm, 5/=l 400mm, S2=λ 61 5mm, f=62.5mm, /?=8mm, λø=0.5μm, and <5/?=0, from Equations (4), (5), and (6) are used, a power series expansion of the phase delays of a /?-type logarithmic asphere or a logarithmic phase plate is found as follows.

Table I

[0065] In Table I, the first row of data is the whole phase delay function of the multifocal lens from Equation (1 ), i.e., .22 (r/R)2- 45.63(r/R)4+0.32(r/R)6+... . The second row of data is the radian phase delay function for an ideal lens arrangement, e.g., Petzval lens, Cooke lens, double Gauss lens, or Cassegrain system. The third row of data is the aberration terms of the phase delay function, which is the difference between the phase delays of the multifocal lens and an ideal lens. The dominant aberration term in the multifocal lens is the third order spherical aberration (i.e., the fourth order term of r/R). For diffraction-limited resolution, the largest allowable O.P.D. is generally 0.25A To achieve a ten-fold increase in the depth of field, the spherical aberration is around 3 wavelengths (i.e., 19.28 λ/2π=3λ), which is a little over ten times the allowable defocus for diffraction-limited imaging. Good performance of our multifocal lens includes spherical aberration in the amount of 1 .8 to 6 wavelengths, while the higher order spherical aberration is largely insignificant. The multifocal lens designed this way will have an extended depth of field from 6 to 10 times that of a conventional ideal lens.

[0066] Another example of our multifocal lens has the same parameter values as above, but with δR/R=0.5 to illustrate the effectiveness of a central obscuration 26. The phase delay for the different terms is shown in the following Table II.

Table

[0067] Although the third order spherical aberration (r/R)4 looks larger than without central obscuration, the effective third order aberration, i.e., the phase delay difference contributed from spherical aberration between the edge of the lens and the edge of the central obscured block is: 25.76- {25.7βx(δR/R)2} =1 9.32 radians. Thus, the effective third order aberration amounts are similar for both the full aperture multifocal lens described by Table I and the centrally obscured multifocal lens described by Table II. Accordingly, the good performance centrally obscured multifocal lens has an effective third order aberration that is still within a range from 1 .8 to 6 wavelengths.

[0068] From the above description, it is apparent that the multifocal lens having an effective third order spherical aberration in the range of 1 .8 to 6 wavelengths, can increase the depth of field six to ten times over that of a conventional lens. This conclusion pertains to any reasonable amount of central obscuration and also is independent of the wavelength of illumination, focal length and best focus object distance.

[0069] The second order term, i.e. (r/R)2 of the series expansion, is not relevant to the increase of depth of field, but has the function of changing the position of center of the focus range. For the second order term, we generally pick a value in a way that the aberration W at the inner edge of aperture 28 and the aberration W at the outer edge of central obscuration 26 have similar values to facilitate the phase plate or lens fabrication. In the case that no central obscuration is used, i.e., δR=0, a coefficient of the second order term is selected so that the aberration W at the edge of aperture 28 is zero.

[0070] There are different ways that these controlled aberrations can be incorporated into the well-known imaging lens arrangements, e.g., Petzval lens or Cooke lens. For an existing lens arrangement, a simple way is to fabricate the aberration part of the multifocal lens as the phase plate 24, which can be attached to the aperture 28 of the lens arrangement. This method is most effective if the aperture 28 of the lens arrangement is outside the last lens element at the image plane (II) side.

[0071 ] Another method of multifocal lens realization is to incorporate the aberration into the lens design of the logarithmic asphere 34. By modifying a surface parameter of the logarithmic asphere 34, the overall phase delay function can still include the ideal lens part and the aberration part. This method has the advantage that no actual lens element is needed, e.g., the flipover of the well-known lens arrangement introduces large amount of spherical aberration, which could be used as the starting design point. Two important features of this embodiment are that it contains good angular resolution as well as good color correction. The desired amount of spherical aberration can also be distributed among multiple lens elements of the design to proved more design flexibility.

[0072] A substantially distance-invariant impulse response is important to the recovery of images having extended depths of focus. A predetermined amount of spherical aberration can be used to produce a more distance-invariant impulse response for effective performance both with and without central obscuration. In the lens with δR=0.S, an optimum amount of spherical aberration has been found to be about 3 waves. However, fairly good image recovery is obtained for a spherical aberration in the range from 1 .8 to 6 waves. Figures 3A to 3F show the effective range for a distance-invariant impulse response. Of note are: 1 ) the width of center peak; 2) the similarity of side lobes; and 3) the energy leaked to side lobes. Figures 4A to 4F show the corresponding impulse responses for a lens with δR=0.

[0073] The above discussions apply also to the case of the κ~type logarithmic asphere. For the yMype logarithmic asphere, the coefficients of the power series for Wchange signs but are otherwise similar, which is shown below by way of example.

[0074] For the κ~type logarithmic asphere, the same design parameters can be used including: 1 61 5mm, f=62.5mm, /?=8mm, λσ=0.5μm, and δR=0. From Equations (4), (7), and (8), a power series expansion of the phase delays of a y-type logarithmic asphere or a logarithmic phase plate is found as shown in Table III.

Table

[0075] As another example of the multifocal lens, Table IV is based on the same parameter values as above, but with δR/R=0.5 to illustrate the effectiveness of a central obscuration 26. This result can be compared to that in Table II.

Table IV

[0076] A difference between the /?-type and γ-Xype phase plates is the sign change for the second and fourth order terms. The fourth order term, which corresponds to third order spherical aberration is positive for the γ~ type lens and negative for the /?-type lens. However, the absolute values of the corresponding third order spherical aberration terms are similar for the same design range.

[0077] To demonstrate the performance of the γ-type lens, Figs. 5A-5F depict the point spread functions for different amounts of third order spherical aberration in units of O. P. D. in combination with a central obscuration of δR/R=0.5. Figs. 6A-6F depict the point-spread functions for different amounts of third order spherical aberration in units of O. P. D with no central obscuration. It is apparent that the width of the point spread function changes from small to large with the object is farther away, which contrasts with the results of the /?-type lens. From the Figs. 5 and 6, it is apparent that the effective spherical aberration is still in the range of between 1 .8 and 6 wavelengths, although the range is positive for the κ~tyPe 'ens and negative for the /?-type lens. As a further comparison between lens types, it is apparent that the /?-type provides a better long distance performance, while the γ-type is favored when the close-in distances are more critical.

[0078] Based on the similarity of the impulse responses over the range of object distances, e.g., to S2, digital processing of the intermediate image 30 can be used to sharpen the images of object points throughout the depth of field. One method of image recovery involves use an inverse filter, such as the Weiner-Helstrom inverse filter. Alternatively, a maximum entropy algorithm can be programmed into the digital processing subsystem, and a preferred approach to the application of this algorithm for image recovery is set forth below.

[0079] For the known, measured, noisy image d, point spread function h, and standard deviation σ^ of noise 9 in ijth pixel of d, and unknown object f, one can write the following relation: d = h **f + & (9) in which the double asterisk (**) is a spatial convolution. For an estimate of the object f, we start with an assumed object fl°> and iterate according to Fig. 5. The maximum entropy criterion is to find an estimate of object /"which will maximize 5 under the constraint of C=Caιm,

where:

(1 0) S = -Y f (In^ 1) tr -J </> (1 1 ) and Caim is the total number of pixels in the image and <f> is the average of image.

[0080] The maximum entropy algorithm is an iterative approach to determining an estimate of object 20. A diagram of the algorithm is shown in Fig. 5, where an unknown object is convolved with the actual point spread function of the lens. Then, noise is added in the process of imaging. Starting with the initial estimate of the object, an image of this object is calculated by convolving with the single point spread function. Then, a difference between the measured blurred image and the calculated blurred image is calculated. If the difference is larger statistically than the noise in the experiment or the criterion of entropy maximization is not reached, the new estimation of the object is generated until both noise constraint and entropy maximization criterion are met, i.e., Equations (10) and (1 1 ) are satisfied.

[0081 ] The single point spread function used in the convolution can be calculated as an average of the point spread functions observed for the different focal depths. However, individual focal distances can be weighted differently to adjust the single point spread function for favoring certain object distances over others for compensating for other effects. The single point spread function could also be varied experimentally to achieve desired results for particular applications or scenes.

[0082] For each iteration, the new estimation of object is calculated from the earlier estimation by adding three (or four) direction-images with appropriate coefficients, i.e., fn+X) = f(n) +δf = fw +∑x,e, (1 2) where flπ+ 1> is the (n+ l)th estimation of object, fn> is the earlier nth estimation, and e, is the Ith direction-image.

[0083] Thus, two key steps of the algorithm are: i) What direction-images e, should be used. ii) How to calculate the corresponding coefficients Xι of the direction images.

[0084] A new metric parameter Tis introduced as a first step to determining the direction-images e,. The parameter r adjusts the pixel values of direction-images derived from a steep ascent method. The parameter r ranges from 0 to 1 , although Y> λ is still possible. When this parameter is larger, more emphasis is given to the larger pixel values in the image, and also there exists more deviation of the direction images e, from direction images derived steepest ascent method. [0085] In the second step, Taylor expansions of 5 and C relative to variables δfare calculated up to second order terms. Hence, the quadratic approximation models St and Ct are established. The quadratic models greatly facilitate the constrained maximization process because these quadratic equations are much easier to solve than the original nonlinear equations in Equations (10) and (1 1 ). The diagram of how to find the next estimation of the object is shown in Fig. 8.

[0086] In order to study the optimum value of metric parameter Y, an extended study has been made of the effect of varying the parameter Y. Three different pictures of varying histograms are used including: binary scene, zebra, and tiger. Each of these pictures has 256x256 pixels with the maximum pixel value scaled to 255. Each picture is blurred using 1 5 normalized impulse responses with the maximum blur consisting of a 5x5 matrix with 1 5 non-zero values and 10 zeros in the outer regions. Gaussian noise is added with a standard deviation σ ranging from 0.2 to 1 .8 in 9 steps. The metric parameter Y\s given 21 values ranging from 0.0 to 1 .0. Hence, in these computer simulations there are about 8,000 cases. It is convenient to use an effectiveness parameter for the number of iterations, which is defined by Lσ/D, where L is the number of loops for the maximum entropy calculation to converge, <τis the noise standard deviation, and D is the number of non-zero pixel in the blurring function. In Fig. 9, we show a plot of Lσ/Dvs. Y, where Tis the metric convergence parameter. The starting images in the algorithm are uniformly gray pictures with pixel values equal to the mean of the blurred images. The rationale for using the parameter, Lσ/D, is explained as follows.

[0087] For the computer simulation, the number of loops L for the maximum entropy recovery is linearly proportional to the area of the point spread function, D, or qualitatively proportional to the severeness of the blur. The loop number is also approximately inversely proportional to the standard deviation of the noise, σ.

[0088] For a wide variety of pictorial content, it is apparent from Fig. 9 that the choice Y=O.4 provides a much faster convergence than T=I . For the method of steepest ascent, T=O, the algorithm does converge but it takes 1 73 times as many loops for a fixed σ/D as compared to the Y=O.4 case (for the zebra). Another feature not immediately apparent from the curves is that when the metric Tis equal to or close to 1 , there is a chance that the algorithm can fail due to stagnation. By the experiment, the recommended value of the metric parameter is from 0.2 to 0.6 and more preferably from 0.3 to 0.5. In summary, it is clear that the use of the metric parameter Tguarantees the convergence and makes the algorithm converge much faster for a wide range of scenes. Stagnation is not observed with 0.3< Y<0.5. This new type of nonlinear digital processing is referred to as a metric parameter - maximum entropy (MPME) algorithm. [0089] A more rigorous mathematical description of the metric parameter -maximum entropy algorithm follows within which control over the metric parameter Tenables the algorithm converge much faster for a wide range of scenes. [0090] Two operators used in this section are defined in the following for convenience: (i) If f \s a vector f = {fj, then f γ is a vector with its components defined as:

(ii) If /"and g are vectors of the same dimension, f = {f,/ a.nd g = {gj, then f.x g is a vector with its components defined as:

[0091 ] To find the solution of {fk} according to the Lagrange multiplier method, a new function is defined as: Q = S-AC (1 3) where is λ is a Lagrange multiplier constant. Now the problem becomes to maximize C? under the constraint C - Cam. Since Q is a function of n variables, where n is the number of pixels in the image, which is usually very large, an iterative numerical method can be used to find the solution. In each iteration, the standard way is first to determine the search directions in which the solution is estimated to lie and then to find the step length along these directions.

[0092] The choice of directional images is important in determining the convergence and speed of the algorithm. In the steepest ascent method, the search direction for maximizing Q is VQ. But in order to adjust the weight of different pixel values fι, the direction can be modified to be: (1 4) [0093] In the above equation, the new metric parameter y improves the speed and reliability of the metric parameter - maximum entropy algorithm. For image deblurring in photography, the larger pixel values will have larger weight, so γ > 0 is chosen to let the algorithm approach the desired larger pixel value faster. Generally, y is chosen from 0 to 1 . When Y = 0, θA becomes the search direction for the steepest ascent method. When y - I, θA becomes the search direction used by Burch et al. in a paper entitled "Image restoration by a powerful maximum entropy method," Comput. Visions Graph. Image Process. 23, 1 1 3-1 28 (1 983), which is hereby incorporated by reference. Neither the steepest ascent method nor the method of Burch et al. incorporate the metric parameter γ , which provides a new mathematical construction that can be manipulated to increase the speed of convergence and avoid stagnation.

[0094] At the maximum point C?, we have: VQ = O (1 5) [0095] This implies VQ- VQ needs to be minimized, too. Accordingly, the next search direction should be Vi V(VQ- VQ) , or VVQ - VQ. Here VVQ is the dyadic gradient whose component is defined as follows: (VV0 =-2_ Q >J dftf, [0096] Again, in order to emphasize the bigger pixel values, the direction is modified to be: (16)

[0097] Substitution of Equation (13) into Equations (14) and (16) yields

the following:

(17)

[0098] Observing the above expression, we know that the two directions actually are linear combinations of many directions, which can be treated as separate search directions, viz.,

(18)

[0099] From Equations (10) and (11 ), we have:

(19)

and

(20) [00100] Substitution of Equations (1 9) and (20) into (1 8) yields the components of each search direction as follows:

(21 ) [00101 ] In the algorithm, e5 and e^ can be disregarded as search directions because when γ = 0.5 they decompose to ei and e2, respectively; and also when γ < 0.5, for small pixel values, they both involve dividing by small numbers, which can cause numerical accuracy problems. Accordingly, ej, e_?, e?, and e^ are chosen as four search directions.

[00102] To simplify the algorithm, three search directions are enough for this problem, we can pick ei, e∑, and ej or we can pick ei, e∑, and e4,- λe3, as another choice. The algorithm converges at about the same speed, although the latter choice is a little better. Here, λ is a constant chosen by Equations (1 3) and (1 5), i.e., λ is given by:

(22) [00103] In either case, three directions can be written as eι, θ2, and β3 for simplicity. In the calculation of search directions, it is apparent that that the directions e^ and ej are basically the convolutions related to the real point spread function, h, and object, f, before being shaped to one dimension. Care needs to be taken to make certain that there is not pixel shift or image position shift after the convolution operation.

[00104] After three search directions are calculated for the current iteration (n), the next task is to find an estimation of the object for the next iteration fn+ 1>, which is defined as:

(23) where δf\s the change of image for the current iteration. It is defined as linear combinations of search directions, with their coefficients to be determined, i.e., δf = x,e, + x2e2 + x3e3 (24) [00105] Since 5 and Care functions of /"that vary in a complicated way, a Taylor expansion can be used to calculate their values as a function of the search directions. Retaining up to the quadratic terms, St and Cf can be

written as follows:

[00106] Substitution of Equation (24) into Equations (25) and (26) yield

the following expressions written in matrix form,

[00107] In Equations (27) and (28), the notation is defined as follows:

where [ -]τ denotes the transpose of matrix. Matrices A, B1 M1 and /V can be

calculated from Equations (20) and (21). [00108] Equations (27) and (28) can be simplified by introducing new variables to diagonalize /?and N. First, the rotation matrix R is found to diagonalize the matrix B, i.e., RBRT = diag(λι,λ2,λi) (30) where diag( ■•) denotes the diagonal matrix.

[00109] A new variable Y is defined as follows: Y = RX (31 ) [001 10] Substitution of Equations (31 ) and (30) into Equations (27) and (28) yield the following expressions:

S1 = S0 + AR }77VY -íY vTT d A.iar g(λ,, A2^)F (32) C1 = C0 + MRTY + -Y7RNR7Y (33) [001 1 1 ] Some eigenvalues of £ may be very small, and this is discussed in the following two cases.

Case i) Assume that none of λ/,λ?, and λ3 is small.

[001 12] We introduce Z, such that:

(34) [001 13] Substituting Equation (34) into equations (32) and (33), yields:

(35) C1 = C0 + MRT diag(

(36)

where:

[001 14] A second rotational matrix I/is introduced to diagonalize P, i.e., VPVτ = dmg(μ,,μ2,μ3) (37)

and also define Ua.s:

U = VZ (38)

[001 1 5] Then, substitution of Equations (37) and (38) into Equations (35)

and (36) yield the following expressions:

(39) C1 diag(μι,μ2,μ2)U.

(40)

[001 16] Combining Equations (31 ), (34), and (38), yields the identity: x-RTdιαs{~k'^iz)VTu (41 )

Case ii) Assume λ3 is small relative to λi and λ2.

[001 1 7] In this case, λ3 ~ 0, y3 - 0 \n Equation (31 ), and also:

(42)

[001 18] Then, Equations (32) and (33) become:

where:

[001 19] A second rotation matrix I/is introduced, such that

(45)

where

[001 20] A new variable U is defined as:

U = VZ (46)

then

(47) μ2,0)U (48) [00121 ] Combining Equations (31 ), (42), and (46),yields the identity:

(49) [00122] The other case, when two values of λi, A?, and λ3 are small, can be treated in a similar way. In general, the following expressions of St and Cr can be written for all cases with the quadratic matrices diagnolized as follows:

(51 ) [00123] The relation between u, and x, (i = 1, 2, 3 or i = 1, 2, or i = 1) can be found in the identities Equations (41 ) or (49).

[00124] Now the maximum entropy problem becomes one to maximize St in Equation (50) under the constraint of Cr = Caim in Equation (51 ). Cf in Equation (51 ) has minimum value of:

(52) [001 25] Clearly, Cmin could be larger than Caim . If this happens, then the maximization of 5 under the constraint of Cf =Ca/m will not have any solution. Accordingly, a new constraint is defined that can always be reached as follows: (53) [00126] The Lagrange multiplier method can be used to solve the maximization in Equation (50) by introducing a new variable Qt as follows:

(54) where Qt corresponds to the Q wave variable at the left side of Equation (54), (a > ^guarantees that the solution found will maximize the entropy instead of minimizing it.

[00127] Substituting Equations (50) and (51) into (54), yields the values of .7/ that maximize Qt as follows:

as. - C1 μ,+a (55) where a\s determined by solving the following equation, which is derived by the substitution of Equation (55) into Equation (51) and by the use of the constraint in Equation (53):

„ .T-, CiS1-C1 1 •r-i .as -c,.2 ~ Q + ∑c,— '■ — L+-∑μX— — '-) =Co (α>0) ^ μ,+a I^ μ,+a (56) [00128] After a is known, coefficients of xi, X2, and X3 are found by Equations (55) and (41) or (49), and the next assumed object for the next iteration can be calculated by Equations (23) and (24). At each iteration, the negative values in the assumed object are reset to zero. [00129] At each iteration if the constraint of C = Caim is satisfied, whether the entropy is maximized is checked by determining if VQ is zero, or whether FS and FC are parallel by calculating the following value: VS VC test = - Vvs-v^Vvc-vc (57) [001 30] The algorithm stops if \ test\ < 0.1 .

[001 31 ] There is a special case when coefficients of xi, X2, and A-j are too large such that the expressions in Equations (25) and (26) are not accurate. The Burch et al. paper deals with this by introducing a distance penalty parameter. However, if the starting estimation of the object is a uniformly gray picture or the blurred picture, then this complexity can generally be avoided. Only when the starting image is random should the extra parameter be introduced in the algorithm but only through the first several loops. A further description of the metric parameter - maximum entropy algorithm is found in a paper authored by the co-inventors entitled "Computational imaging with the logarithmic asphere: theory" J. Opt. Soc. Am. A/Vol. 20, No. 1 2, December 2003, which is hereby incorporated by reference.

[001 32] In addition to increasing the speed of convergence and avoiding stagnation improved deblurring and image recovery are possible. The metric parameter-maximum entropy algorithm or MPME algorithm improves image quality by increasing the contrast of the recovered image. Adjustments to the metric parameter y, particularly to within the range of 0.2 to 0.6 result in a modulation transfer function having a more rectangular form, which preserves contrast of higher spatial frequency components. The effect of the metric parameter γ \s also evident on the point-spread function as a reduction in side lobe oscillations apparent in the intermediate image. The final point images are closer to true points with little or no ringing. Disappearance of the oscillating rings also increases contrast.

[001 33] The MPME algorithm provides an iterative digital deconvolution method capable of starting with any image. An estimate of the next new image can contain a linear combination of directional images. The metric parameter ^ modifies the directional images from those provided by conventional maximum entropy algorithms, while reconciling the directional images of the conventional algorithms as integer instances of the metric parameter γ. Preferably, a quadratic Taylor expansion is used to calculate the values of the entropy 5 and the statistical noise constraint Cas functions of the search directions. The modified statistical noise constraint assures an iterative solution of the new image estimate.

[00134] The metric parameter-maximum entropy algorithm (MPME) has an important range of applications due to the "box-like" form of the resulting overall modulation transfer function, as shown in Fig. 1 5. Hence, at high spatial frequencies the contrast of any digital image will be higher than is typical of the classical fall-off of photographic images. As is well known, the incoherent image optical transfer function falls off in a triangular-like manner as the spatial frequency ranges from zero to the Nyquist limit, also known as the cutoff frequency. The MPME algorithm provides sharper, crisper, high contrast output. While there are earlier algorithms that provide some form of high-frequency or edge sharpening, these earlier algorithms amplify the high-frequency noise as well. For example one such algorithm consists of a two-dimensional FFT, followed by high-frequency emphasis, followed by an inverse FFT. However, as is well known, these earlier methods of providing emphasis or higher contrast at the higher spatial frequencies amplify the noise in the image. From experiments using the MPME algorithm, as is apparent from a study of the operation as shown in Figs. 7 and 8, the MPME algorithm does not have this drawback. The MPME algorithm provides sharper contrast at the higher spatial frequencies without undue amplification of the noise. Therefore, this contributes to its general applicability in digital image processing.

[001 35] The optional central obscuration 26 of the aperture 28 as apparent in Fig. 2 has been introduced to improve the system performance. By incorporating the central obscuration 26 into the multifocal imaging systeml 2, the processed image can have higher resolution and higher contrast, especially for the close in object distance. The details of this comparison are described in the following paragraphs.

[001 36] To illustrate the improved performance introduced by the centrally obscured logarithmic asphere, Figs. 10 and 1 1 compare the images produced through a full aperture logarithmic asphere with the images produced through a centrally obscured logarithmic asphere. The imaging of a two-point object simulation is shown in Fig. 10, based on a full aperture ^-design logarithmic asphere. The design parameters are: 5/=l 400mm close distance, 52=1 61 5mm far distance, and sø=l 500mm center distance with an f/# of the lens at 4. The diffraction-limited depth of field is ±8mm. The two object points are separated by the diffraction limit distance, viz., 2.27μm. The images of the object blurred by the logarithmic asphere are shown in Figs. 10(b)-(d) for the object distances s=l 450mm, 1 500mm, and 1 580mm, respectively. Figs. 10(e)-(h) show the maximum entropy recovery results for the images in Figs. 10(a)-(d), respectively. For comparison, the performance of an idealized lens is shown in Figs. 10(a) and 10(e). Immediately apparent is the excellent recovery due to the deconvolution inherent in the maximum entropy algorithm. A standard deviation of the noise used in these recoveries is σ=λ .5. The point spread function used for Fig. 10(b)-(d) is the average point spread function of the logarithmic asphere over the design range, and the point spread function used for Fig. 10(a) is the actual diffraction blur of the ideal lens. In the maximum entropy recovery using any single member of the impulse responses will lead to near perfect recovery at that distance. However, for extended depth-of-field, it is more advantageous to use the averaged point spread function. Still, as seen in Figs. 10(b)-(d), by the elimination of most of the oscillation rings and the narrowing of the blurs, nearly diffraction-limited resolution can be achieved over the entire range. Faint rings are apparent in the recoveries shown in Figs, (g) and (h).

[001 37] A similar simulation using the centrally obscured logarithmic asphere 26 is shown in Figs. 1 1 (a)-(o). The same close (s/=l 400mm), far (52= 1 61 5mm) and center (5ø= l 500mm) distances are used. The center obscuration <5/?=0.5. i.e., by area 25% of the aperture is not used. This obscuration causes an approximately 25% loss of light intensity, which needs to be considered in a final design trade-off. In Fig. 1 1 , the rows are for object distances 1450mm, 1492mm, 1 500mm, 1 508mm, and 1 580mm, respectively, and the columns are for ideal images, intermediate (blurred) images, and recovered images, respectively. The conventional depth of field (for a full aperture f/4.0) ranges from 1492mm to 1 508mm. A single average impulse response over the design range is used for all five recoveries. The similarity of blur for the logarithmic asphere is clearly seen from the center column of the intermediate (blurred) images, all have two bright peaks at the center accompanied by low intensity oscillating rings. The center bright peaks also have similar sizes. The oscillating rings do not pose a problem since excellent recoveries are achieved for all five images. The two-point recovery (right column) shows excellent diffraction-limited resolution from 1450mm to 1 580mm. At s=l 420mm (not shown), the two points are still easily resolvable, but not as clearly as those in the figures.

[001 38] The performance improvements, particularly for close-in distances (β-design), is believed due to both the narrower central peak of the average point spread function and the similar oscillating ring structure of the point spread function over the designed object range. These two factors lead to a point spread function that varies less with object distance 5, so that the average point spread function used in the digital processing can provide a significantly improved output. Thus, in comparing Figs. 10 and 1 1 , a significantly improved result can be achieved by using the centrally obscured logarithmic asphere. The rings are more suppressed and the two points are more clearly resolvable over a range of object distances. In addition, the central obscuration can avoid conditions such as contrast inversion for larger amounts of defocus.

[001 39] To further demonstrate the improved performance of centrally obscured logarithmic asphere, we show in Figs. 1 2 and 1 3 intensity distributions associated with imaging the same two-point object using three logarithmic aspheres having different central obscuration values of (5/?=0, δR=03, and δR=0.S, respectively. When the object is at a distance of s=l 450mm, or 50mm closer than the best focus plane, the recovered one dimensional pixel values in the two point direction are plotted in Fig. 1 2 along with the pixel values of the blurred image by an ideal lens. For producing the ideal lens curve, the full aperture of the lens is used, i.e., <5/?=0. A fully resolvable two point recovery is produced with center obscuration of δR-0.5. However, when the obscuration is reduced to δR=0.3, the recovery points are barely resolvable; and when there is no central obscuration, δR=0, the recovery points are unresolvable. These curves clearly show the performance improvement made by a correctly sized central obscuration. The increase in performance is believed mainly due to the similarities of the point spread functions over the design range of object distances instead of from the increase of the depth of field by central obscuration, since we have seen from the earlier section that the increase in depth of field by central obscuration of aperture is, by itself, much more limited. Fig. 1 3 illustrates a set of similar curves for an object is at a distance of 5=1 500mm, the best focus position for the logarithmic aspheres and the ideal lens. Again, advantages of the logarithmic asphere with central obscuration for both the resolving power and contrast are apparent.

[00140] The improved performance made possible by the invention has particular benefits for photography, which can be observed by comparison from the pictures of Figs. 14(a)-(d). In this simulation, the tiger picture is at a distance of 1 580mm (i.e., 80mm beyond the best focus position). Fig. 14(a) is the recovered image for the logarithmic asphere without center obscuration. Figs. 14(b) and 14(d) are the recovered images for the logarithmic asphere with central obscuration of <5/?=0.3 and δR=0.5, respectively. Fig. 14(c) shows the tiger image reproduced by an ideal lens with full aperture for comparison purposes. The logarithmic aspheres both with and without obscuration are capable of extending the depth of field. However, the recovered images for logarithmic aspheres with obscuration are better because there are fewer artifacts. The artifacts of the recovery are believed to appear because of differences between the point spread functions through the range of object distances, while the average point spread function over the design range is used for all the recoveries. In inspecting these pictures, it is of note that the width of tiger whiskers in the simulation is about 0.7μm, which is smaller than the diffraction-limited spot size. [00141 ] In intensity imaging systems, it is common to characterize their performance by an optical transfer function. Extending this notion to a computational imaging system, in principle, the overall frequency response can be found by dividing the spectrum of the recovered image by that of the input object. Thus, to find the overall frequency response, the images of a point source can be calculated at various object distances, and the maximum entropy algorithm can be applied to these intermediate images to recover the point object. The recoveries can be considered as the combined impulse response of the integrated computational imaging system. A Fourier transform of the recoveries is plotted in Fig. 1 5. The curves are the combined transfer functions of the system over a range of object distances. The transfer function of the system is circularly symmetric, and Fig. 1 5 shows its values along the radial direction over a range of focal depths. The relative spatial frequency 1 .0 corresponds to a cutoff frequency of the diffraction-limited lens for the same imaging settings. The amplitude of the overall transfer function of the new system is increased to the diffraction limit over an extended object range. In addition, the phase of the overall transfer function is zero due to the circular symmetry of the impulse response. The diffraction-limited performance for the integrated computational imaging system over an extended depth of field is clearly seen from these curves.

[00142] However, the concept of overall transfer function is only an approximate index of the system performance because of the nonlinear digital processing involved. In other words, different overall transfer functions can be expected for various objects. Nonetheless, the transfer function shown in Fig. 1 5 is a good indication of performance of the integrated imaging system. Fig. 1 6 references the overall transfer function for objects at a distance of 5=1 580mm, where transfer functions for both a point and an edge object are plotted. The recoveries from the blur images by the logarithmic lens provide the overall point spread function and overall edge response function, respectively. From Fig. 16, it is apparent that the overall transfer functions are similar for these two cases.

[00143] From the foregoing description, it will be apparent that an improved system, method and apparatus for imaging are provided using a multifocal imaging system in which spherical aberration is the predominate form of blur for intermediate imaging and in which a central obscuration can be used for making more intermediate images more susceptible to correction by digital processing for increasing both resolution and depth of field. Variations and modifications in the herein described system, method, and apparatus will undoubtedly become apparent to those skilled in the art within the overall spirit and scope of the invention.

**Previous Patent:**[3.2.0] HETEROCYCLIC COMPOUNDS AND METHODS OF USING THE SAME

**Next Patent: PROBE-STYLE QUICK-ATTACH INTERCONNECT MECHANISM**