Field

The present invention relates to a method of watermarking volumetric images, more particularly but not exclusively, volumetric medical images.

Background

Digital medical imaging has gained importance over the years and has become the primary medium for medical imaging in the healthcare industry. In parallel with the advent of digital medical imaging, rapid development of information and communication technology ( ICT) has facilitated the advancement of teleradiology by enabling faster and wider access to electronic medical records, including medical images. With increasing demand by medical experts at different locations to study medical images from people all over the globe in teleradiology, medical images are increasingly transmitted through the open network (e.g. the internet). When medical images which contain sensitivity information of patients are transmitted over the open network, they become vulnerable to corruption by noisy transmission channels, or attacks by hackers or individuals with malicious intends. The malicious attacks may include obtaining private information about the patient, changing patient information in the image header, tampering the image pixel content, and the like. These attacks will cause the images to lose integrity. The use of these tampered medical images could potentially lead to wrong diagnosis of diseases, which could result in an erroneous treatment for the patient, or misjudged insurance claims. In the worst case, it might even result in death of patients. Consequently, the security of digital medical images becomes an important problem that needs to be addressed.

The Digital Imaging and Communication in. Medicine (DICOM) standard (used for distributing and viewing medical images) defines a set of security profiles that application entities should conform to during the exchange of medical data. These profiles include secure use profiles, secure transport connection profiles, digital signature profiles and media storage security profiles. A DICOM image file consists of a header that stores patient's information and image data. However, the authenticity of DICOM medical images can be compromised as it is possible for a hacker to change the header of another person's DICOM image file to his own in order to, for example, claim monetary gain from insurance company. To counter this problem, DICOM proposes to verify the authenticity of medical images by using digital signature. However, digital signature relies on the inherent strength of cryptographic hash function and, more importantly, does not provide tamper localization capability.

Digital watermarking is a data hiding technique that has the potential to protect medical images, even after the images leave the network. A number of digital watermarking techniques for medical images have been reported and these techniques have shown encouraging results for medical image integrity control. The advantages of using watermarking include (i) enabling authentication information such as the metadata to be embedded into the medical images as visually unperceivable watermark payload, (ii) providing data integrity protection beyond the point of internal network, and (iii) enabling tamper localization that enables the recipient to detect whether the image has been tampered. The ability to locate the tampered regions enables the recipient to determine the severity of the tamper. If the tampered region is unimportant medically, then the recipient has an option to continue using the medical image without having to request for retransmission (which saves time and cost).

Existing watermarking techniques for addressing integrity control and authenticity verification of DICOM medical images, however, generally focus on images acquired by two-dimensional (i.e. "2-D") imaging modalities such as X- ray radiography, or directly apply 2-D techniques on individual image of three- dimensional (i.e. "3-D") data acquired. by imaging modalities such as computed tomography (CT) and magnetic resonance imaging (MRI). However, as 3-D imaging modalities often produce a large number of cross-sectional image slices for a volumetric set of interest (e.g. over 100 images for a knee MRI dataset), direct application of 2-D techniques on 3-D volumetric data is thus computationally inefficient and prohibitively time consuming.

One object of the present invention is therefore to. address at least one of the problems of the prior art and/or to provide a choice that is useful in the art.

· ^■ Summary

According to a 1 ^st aspect of the invention, there is provided a method of embedding data for use as watermark for a volumetric image. The method comprises defining a watermark embedding threshold as a function of bit depth of the volumetric image, determining which pixels in the volumetric image are to be used for watermarking with reference to the watermark embedding threshold, and embedding the data in the determined pixels for watermarking the volumetric image. An advantage of the described embodiment is that with the watermark embedding threshold, which is adaptive to the bit depth of the image, provides sufficient embedding capacity regardless of the bit depth of an image to be watermarked. Preferably, the volumetric image may be configured in DICOM format, and the adaptive embedding threshold may be configured according to the equation, 0=2 ^P"7 , wherein Q is the adaptive embedding threshold, and p is the bit depth of the volumetric image. Further, the data to be embedded may include metadata corresponding to header information of the volumetric image. More preferably, the data to be embedded may include a hash code generated with reference to the volumetric image using the Secure Hash Algorithm-256.

Yet further, determining which pixels in the volumetric image are to be used for watermarking may preferably include dividing the volumetric image into non- overlapping pixel blocks each of resolution size 2x2 pixels, and processing each pixel block to embed the data until entire bit stream of the data is embedded. More preferably, processing each pixel block may comprise selecting a reference pixel having a reference pixel value from the pixel block, wherein the remaining pixels in the pixel block are subsequently defined as alterable pixels, sequentially selecting respective alterable pixels having corresponding pixel values, determining respective absolute values of the difference between the reference pixel value and each pixel value of the respective alterable pixels, and comparing the respective absolute values with the adaptive embedding threshold for updating the corresponding pixel values of the respective alterable pixels in accordance with a set of predefined rules. ln addition, the set of predefined rules may include, if an absolute value is less than the adaptive embedding threshold, updating a pixel value corresponding to the sum of the pixel value and adaptive embedding threshold if a data bit to be embedded has a value of one, or updating a pixel value corresponding to a difference obtained by subtracting the pixel value from the adaptive embedding threshold if a data bit to be embedded has a value of zero.

Furthermore, the set of predefined rules may also preferably include, if an absolute value is greater or equal to the adaptive embedding threshold, updating a pixel value corresponding to the sum of the pixel value and adaptive embedding threshold if the pixel value is greater or equal to the sum of the reference pixel value and adaptive embedding threshold, or updating a pixel value corresponding to a difference obtained by subtracting the pixel value from the adaptive embedding threshold if the pixel value is less than the sum of the reference pixel value and adaptive embedding threshold.

Yet preferably, the method may further comprise generating a location signal which includes the respective locations of the selected reference pixels of corresponding pixel blocks, and encrypting the location signal using public key cryptography. The data to be embedded may further include a CRC-32 code generated with reference to the unencrypted location signal. Preferably, the method may also further comprise obtaining a first watermark layer corresponding to the data embedded in the determined pixels, and including a second watermark layer in the volumetric image for tamper localization.

According to a 2 ^nd aspect of the invention, there is provided an apparatus for embedding data for use as watermark for a volumetric image, configured to perform the method according to the 1 ^st aspect of the invention.

According to a 3 ^rd aspect of the invention, there is provided a method of watermarking a volumetric image, which includes a leading image and a plurality of non-leading images. The method comprises generating respective sets of horizontal error detection codes corresponding to image characteristics of pixel blocks of each non-leading image and a set of vertical error detection codes corresponding to image characteristics obtained by summing respective pixel blocks of the leading image and subsequent non-leading images arranged at same horizontal coordinates, and embedding the set of vertical error detection codes in the leading image and respective sets of horizontal error detection codes in corresponding non-leading images for watermarking the volumetric image.

Beneficially, the described embodiment, according to the 3 ^rd aspect of the invention, is able to reduce processing time incurred during de-watermarking (since a check may first be carried out using the vertical error detection codes and if there is no error, there is no need to check the horizontal error detection codes, thus, avoiding the need to perform checks on multiple horizontal error detection codes for non-leading images in tamper-free situations), and to improve the tamper localization resolution by exploiting the characteristics of multiple image slices contained within a watermarked volumetric image. Additionally, the method may allow for detection of missing image slices using the vertical error detection codes, and also able to provide a smaller area (and hence better spatial resolution) for tamper localization display. Preferably, the method may further comprise defining an adaptive embedding threshold as a function of bit depth of the volumetric image, and wherein embedding the set of vertical error detection codes and respective sets of horizontal error detection codes is based on the adaptive embedding threshold. Yet preferably, the method may also further comprise creating a watermark layer corresponding to the embedded set of vertical error detection codes and respective sets of horizontal error detection codes, and a further watermark layer in the volumetric image for authenticity verification.

Additionally, the leading image may be the first image in the volumetric image, and the non-leading images may be images sequentially arranged and subsequent to the first image in the volumetric image. On the other hand, the leading image may also be any image as selected in the volumetric image, and the non-leading images may then be the remaining images, excluding the selected image, in the volumetric image. More preferably, generating the respective sets of horizontal error detection codes may further comprise defining a first threshold, wherein the first threshold is twice the value of the watermark embedding threshold, dividing respective non-leading images into non-overlapping pixel blocks each of resolution size 18x18 pixels, and processing each pixel block of the respective non-leading images to embed the corresponding set of horizontal error detection codes until entire bit stream of the corresponding set of horizontal error detection codes is embedded in the associated non-leading images. Further preferably, processing each pixel block may comprise calculating a CRC-16 code of the pixel block, dividing the pixel block into smaller non- overlapping pixel blocks each , of resolution size 3x3 pixels, and processing each smaller pixel block to embed the CRC-16 code until entire bit stream of the CRC- 16 code is embedded. More specifically, processing each smaller pixel block may comprise selecting a central pixel as a reference pixel having a reference pixel value from the pixel block, wherein the remaining the pixels in the pixel block are subsequently defined as alterable pixels, sequentially selecting respective alterable pixels having corresponding pixel values, determining respective absolute values of the difference between the reference pixel value and each pixel value of the respective alterable pixels, and comparing the respective absolute values with the first threshold for updating the corresponding pixel values of the respective alterable pixels in accordance with a set of predefined rules. The set of predefined rules may include, if an absolute value is less than the first threshold, updating a pixel value corresponding to the sum of the pixel value and first threshold if a data bit to be embedded has a value of one, or updating a pixel value corresponding to a difference obtained by subtracting the pixel value from the first threshold if a data bit to be embedded has a value of zero. Furthermore, the set of predefined rules may include, if an absolute value is greater or equal to the first threshold, updating a pixel value corresponding to the sum of the pixel value and first threshold if the pixel value is greater or equal to the sum of the reference pixel value and first threshold, or updating a pixel value corresponding to a difference obtained by subtracting the pixel value from the first threshold if the pixel value is less than the sum of the reference pixel value and first threshold.

Yet preferably, generating the set of vertical error detection codes may further comprise defining a second threshold, wherein the second threshold , is eight times the value of the adaptive embedding threshold, dividing the leading image into non-overlapping pixel blocks each of resolution size 1 8x18 pixels, and processing each pixel block to embed the set of vertical error detection codes until entire bit stream of the set of vertical error detection codes is embedded in the leading image.

Specifically, processing each pixel block may preferably comprise calculating a CRC-16 code of the pixel block, dividing the pixel block into four non-overlapping pixel blocks each of resolution size 9x9 pixels, calculating a CRC-8 code of respective vertical 9x9 pixel blocks, wherein a vertical 9x9 pixel block comprises a sum of each of the four 9x9 pixel blocks of the leading image with respective plurality of 9x9 pixel blocks of subsequent non-leading images arranged at same horizontal coordinates, forming resultant codes from respective CRC-8 codes and CRC-16 the code, wherein each resultant code comprises four data bits of the CRC-16 code and the respective CRC-8 codes, and processing each of the four 9x9 pixel blocks to embed the corresponding resultant code until entire bit stream of the resultant code is embedded.

More preferably, processing each of the four 9x9 pixel blocks may comprise dividing the 9x9 pixel block into smaller non-overlapping pixel blocks each of resolution size 3x3 pixels, and processing each smaller pixel block to embed the resultant code until entire bit stream of the resultant code is embedded in the 9x9 pixel block. Particularly, processing each pixel block may comprise selecting a central pixel as a reference pixel having a reference pixel value from the pixel block, wherein the remaining the pixels in the pixel block are subsequently defined as alterable pixels, sequentially selecting respective alterable pixels having corresponding pixel values, determining respective absolute values of the difference between the reference pixel value and each pixel value of the respective alterable pixels, and comparing the respective absolute values with the second threshold for updating the corresponding pixel values of the respective alterable pixels in accordance with a set of predefined rules.

The set of predefined rules may include, if an absolute value is less than the second threshold, updating a pixel value corresponding to the sum of the pixel value and second threshold if a data bit to be embedded has a value of one, or updating a pixel value corresponding to a difference by subtracting the pixel value from the second threshold if a data bit to be embedded has a value of zero. Yet preferably, the set of predefined rules may also include, if an absolute value is greater or equal to the second threshold, updating a pixel value corresponding to the sum of the pixel value and second threshold if the pixel value is greater or equal to the sum of the reference pixel value and second threshold, or updating a pixel value corresponding to a difference by subtracting the pixel value from the second threshold if the pixel value is less than the sum of the reference pixel value and second threshold.

According to a 4 ^th aspect of the invention, there is provided an apparatus for watermarking a volumetric image, configured to perform the method according to the 3 ^rd aspect of the invention.

' ^'

According to a 5 ^th aspect of the invention, there is provided a method for de- - watermarking a watermarked volumetric image, the watermarked volumetric image including a leading image and a plurality of non-leading images. The method comprises processing the leading image to extract a corresponding set of error detection codes; the error detection codes including sets of horizontal error detection codes corresponding to image characteristics of pixel blocks of each non-leading image and a set of vertical error detection codes corresponding to image characteristics obtained by summing respective pixel blocks of the leading image and subsequent non-leading images arranged at same horizontal coordinates, and determining if the watermarked volumetric image is tampered based on the extracted set of error detection codes. If tampering is determined, an area on the watermarked volumetric image corresponding to occurrence of the tampering is identified based on the extracted set of error detection codes. Preferably, the method may further comprise dividing the leading image into a plurality of non-overlapping 18x18 pixel blocks, wherein processing the leading image includes processing each pixel block. More specifically, processing each pixel block to extract the corresponding set of codes, may further comprise dividing each pixel block into four non-overlapping 9x9 pixel blocks, processing each 9x9 pixel block to extract respective resultant codes, and retrieving a CRC- 16 code and four sets of CRC-8 codes from the extracted resultant codes of all four 9x9 pixel blocks. Further preferably, processing each 9x9 pixel block may comprise dividing the 9x9 pixel block into non-overlapping 3x3 pixel blocks, and processing all the 3x3 pixel blocks to obtain a resultant code of the 9x9 pixel block. Particularly, processing each 3x3 pixel block may comprise selecting a central pixel as a reference pixel having a reference pixel value from the pixel block, wherein the remaining pixels in the pixel block are subsequently defined as alterable pixels, sequentially selecting respective alterable pixels having corresponding pixel values and comparing with the reference pixel value for updating the respective pixel values in accordance with a first set of predefined rules, determining respective absolute values of the difference between the reference pixel value and each updated pixel value of the respective alterable pixels, and comparing the respective absolute values with a threshold for determining the type of data bit to be extracted in accordance with a second set of predefined rules. All the extracted data bits in combination form the resultant code and the threshold is eight times the value of an adaptive embedding threshold which may be defined to be a function of bit depth of the watermarked volumetric image.

The first set of predefined rules may include, if the pixel value is less than the reference pixel value, updating the pixel value corresponding to the sum of the pixel value and threshold, or if the pixel value is greater than the reference pixel value, updating the pixel value corresponding to a difference by subtracting the pixel value from the threshold. The second set of predefined rules may include, if the pixel value is greater than the reference pixel value and an absolute value is less than the threshold, a data bit with a value of one is extracted, or if the pixel value is less than the reference pixel value and an absolute value is less than the threshold, a data bit with a value of zero is extracted. Further more preferably, in determining if the watermarked volumetric image is tampered, the method may further comprise determining if the respective extracted resultant codes is each of a length of twelve bits or if the four sets of retrieved CRC-8 codes matches corresponding CRC-8 codes calculated using the respective four 9x9 pixel blocks, and matching the retrieved CRC-16 code with a CRC-16 code calculated using the leading image to determine if the corresponding 18x18 pixel block of the leading image is tampered. According to a 6 ^th aspect of the invention, there is provided an apparatus for de- watermarking a watermarked volumetric image, configured to perform the method according to the 5 ^th aspect of the invention.

According to a 7 ^th aspect of the invention, there is provided a method of watermarking a volumetric image using first and second watermark layers, which includes a leading image and a plurality of non-leading images. The method comprises embedding data in the volumetric image with reference to a watermark embedding threshold which is dependent on bit depth of the volumetric image to obtain the first watermark layer, generating respective sets of horizontal error detection codes corresponding to image characteristics of pixel blocks of each non-leading image and a set of vertical error detection codes corresponding to image characteristics obtained by summing respective pixel blocks of the leading image and subsequent non-leading images arranged at same horizontal coordinates, and embedding the set of vertical error detection codes in the leading image and respective sets of horizontal error detection codes in corresponding non-leading images to obtain the second watermark layer.

It should be apparent that features relating to one aspect of the invention may also be applicable to the other aspects of the invention.

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. Brief Description of the Drawings

Embodiments of the invention are disclosed hereinafter with reference to the accompanying drawings, in which:

Figure 1 is a component diagram of a reversible 3-D watermarking method, according to a first embodiment of the invention;

Figure 2 is a flow diagram of a method for performing authenticity watermarking, as shown in Figure 1 , at a sender's end;

Figure 3 is a flow diagram of a method for data embedding (i.e. watermarking), which is utilised in the method of Figure 2;

Figure 4 is a flow diagram of a method for data extraction (i.e. de-watermarking), which is applied to a watermarked image at a receiver's end;

Figure 5 shows an embedding structure in relation to a leading image, and a series of non-leading images for integrity watermarking, as shown in Figure 1 ;

Figure 6 is a flow diagram of a method for integrity watermark embedding for the series of non-leading images of Figure 5;

Figure 7 is a flow diagram of a method for integrity watermark embedding for the leading image of Figure 5;

Figure 8 is a flow diagram of a method for detecting and localizing tampered areas of the watermarked image at the receiver's end, as part of a process for restoring the original image;

Figure 9 is a flow diagram of a method for carrying out tampering examination, which is utilised in the method of Figure 8;

Figure 1 0 is a table depicting parameters of test DICOM image sets used for performance evaluation of the 3-D watermarking method of Figure 1 ;

Figures 1 1 a and 1 1 b are screenshots of evaluation results corresponding to successful watermarking using MRI heart dataset, and successful de- watermarkirig using the same dataset respectively;

Figure 2 shows screenshots of evaluation results corresponding to successful tamper detection and localization, with the identified tampered image slices 1 , 2, 5, and 8 shown enlarged, and the tampered areas are indicated in red boxes; Figure 3 is a table comparing the respective time required to watermark DICOM images using a prior art method, and the 3-D watermarking method of Figure 1 ; Figure 14 is a table comparing the respective time required to de-watermark DICOM images using the prior art method, and the 3-D watermarking method of Figure 1 ; Figure 15 are screenshots of evaluation results relating to systematic tampering localization;

Figures 16a and 16b are respective screenshots of evaluation results relating to the resolution of tamper localization using the prior art method, and the 3-D watermarking method of Figure 1 , when applied to systematic tampering;

Figures 17a to 17c are screenshots of evaluation results relating to real-case tampering localization; and

Figures 18a and 18b are respective screenshots of evaluation results relating to the resolution of tamper localization using the prior art method, and the 3-D watermarking method of Figure 1 , when applied to real-case tampering.

Detailed Description of Preferred Embodiments

1. Methodology

Teleradiology brings convenience of global medical record access, along with concerns over security of medical images transmitted over the open network. With increasing adoption of 3-D imaging modalities, it is vital that security mechanisms designed to protect them must be able to handle large volume of medical images efficiently. With the above in mind, a reversible 3-D watermarking method 100 for watermarking and de-watermarking volumetric medical images, according to an embodiment shown in the component diagram of Figure 1 , is disclosed.

In the current embodiment, the 3-D watermarking method 100 is particularly configured to implement dual layer reversible watermarking, by employing an _, authenticity watermarking method 102 (for creating a first watermark layer) and an integrity watermarking method 104 (for creating a second watermark layer). The authenticity watermarking method 1 02 uses an adaptive embedding threshold coupled with public key encryption to insert patient information and relevant authentication data as a first watermark layer, whereas the integrity embedding method 1 04 inserts tamper localization data as a second watermark layer. Specifically, the integrity embedding method 104 employs a technique that inserts the tamper detection watermark using two different ways intended for leading image and non-leading images respectively, which essentially embeds each zoned block of the leading image in the volumetric image set with both horizontal CRC (cyclic redundancy check) and vertical CRC, and that of other non-leading images with only horizontal CRC.

In other words, the integrity embedding method 1 04 comprises a method for data embedding for non-leading images 1 06, and a method for data embedding for leading image 1 08. One of the objects of the 3-D watermarking method 1 00 is to enable achievement of a significant improvement in computational efficiency, and also to obtain better resolutions for examining the tampered areas in order to provide a more practical solution for protecting 3-D volumetric images, when they are relayed through the open network to be shared with other collaborating medical professionals.

2. Authenticity Watermarking (as the First Watermark Layer)

Figure 2 is a flow diagram of the authenticity watermarking method 1 02, which is configured to be performed on any DICOM images at a sender's end, during the data embedding stage before transmission . Accordingly, at step S1 022, an adaptive first watermark embedding threshold, "Q _? ", is defined to ensure that there is sufficient embedding capacity regardless of the bit depth of an image to be watermarked. It is to be appreciated that the image has a resolution of WxH pixels. Hence, "Q " is therefore made a function of the bit depth "p" of the image, where "Q ₁ = V ¹ " , instead of a fixed value. It should be appreciated that the choice of "Q " used depends on the correlation of neighbouring pixel values. As the bit depth of the image increases, the difference in value between neighbouring pixels also increases proportionally. Therefore, "Q ", as defined, caters to images with bit depth larger than eight-bits to ensure there is enough embedding capacity and invisibility. In this respect, it will be appreciated that medical images are usually twelve-bits large. Moreover, it has been shown through a study that the minimum value for "Or" for an eight-bit image is to be two, and therefore the formulation of "Q ₁ = 2 ^Ρ'Ί " is adopted.

At step S1 024, the image is divided into non-overlapping square blocks of size "Ν ^χ Ν', where 1 ≤ N≤ min ( W, H) . In this instance, "ΛΓ is defined with a value of "2". It is also to be appreciated that in cases where " W or "H" is not a multiple of "ΛΓ, the excess remaining part at the end of each block row or block column of the divided image are to be appended to the block (of size "Λ/χΛΓ which is obtained after dividing the image into the non-overlapping square blocks) to its left or above, respectively.

As the purpose of carrying out authenticity watermarking is to embed and keep the metadata (e.g. patient's information from DICOM header) _, confidential, a reference pixel location, of respective 2x2 pixel blocks which is used during the data embedding as further elaborated in Figure 3, is kept secret so that the security of the embedded metadata is guaranteed. In this embodiment, public- key cryptography (i.e. the Rivest-Shamir-Adleman encryption algorithm (RSA)) is employed to protect the reference pixel location information. Specifically, a location signal that denotes the locations of all randomly selected reference pixels is generated. The generated location signal is particularly obtained by concatenating all the locations of the randomly selected reference pixels into a single string. At step S1026, this location signal is then encrypted using the public key of the intended recipient. Moreover, a CRC-32 code, which is used in the IEEE 802.3 standard, of the location signal is also calculated for the purpose of certifying the correctness of the decrypted location signal at the receiver's end. Additionally, for the purpose of verifying successful recovery of the original image, a Secure Hash Algorithm-256 (SHA-256) hash code of the original image is calculated. ' At step S1 028, the authentication data comprising the collation of the hash code, together with the encrypted location signal and the CRC of the location signal, is formed. This authentication data is subsequently embedded as the first watermark layer at step S1 021 0, in which each 2x2 pixel block is processed in a raster scan manner until the entire authentication data bit stream is embedded in the image. Depending on the embedding capacity available, embedding of the authentication data may be terminated at certain blocks when the entire stream of authentication data is considered to be fully embedded. As a result, it will therefore be apparent that the lower part of the image may not contain the authentication data, since there is not repeatedly embedding of the authentication data once the initial string has been completely embedded.

The embedding of the authentication data performed at step S1 0210 is now detailed in Figure 3, which depicts a corresponding flow diagram of a data embedding method 300. Referring to Figure 3, at step S302, a pixel from the currently being processed 2x2 pixel block is first randomly selected and thereafter defined as reference pixel, in which the pixel value "R" remains unchanged throughout the data embedding process. The other pixels in the current block are then defined as alterable pixels whose pixel values As are to be subsequently modified. At step S304, a first alterable pixel value "A" is selected, and the value of "\R-A\" is computed, and compared with "Q" at step S306 to obtain a result. In this instance, it is to be appreciated that "O = Or". The result is then checked at step S308 to determine a next step to be taken in the flow of the data embedding method 300. If the result indicates that "\R-A\ < Q", the alterable pixel is then determined to be able to carry a bit and is consequently updated as follows at step S310: (a). If the bit to be embedded is "1 ", the value of "A" is modified to "A+Q"; (b). Otherwise, if the bit to be embedded is "0", the value of "A" is then modified to

-A-cr.

On the other hand, if the result indicates that "\R-A\≥Q the alterable pixel is conversely determined as unable to carry a bit, but "A" is nonetheless updated as follows at step S.312: (a). If " A≥ R + Q ", "A" is modified to a value of "A+Q"; (b). Otherwise, if " A < R + Q ", "A" is then modified to a value of "A-Q". Regardless of either step S310 or step S312 is taken, the data embedding method 300 eventually routes to step S314, in which steps S306 to S312 are thereafter to be repeated for all other remaining alterable pixels in the current block. Once all the alterable pixels have being processed, data embedding for the current 2x2 pixel block is deemed to be completed.

It is to be appreciated that the description above for the data embedding method 300 applies in relation to data embedding being performed in respect of a single pixel block, and is further to be understood that the data embedding method 300 is also subsequently applied to for all the 2x2 pixel blocks in the image to be watermarked, as afore mentioned in step S10210 of Figure 2.

With respect to Figure 4, a flow diagram of a data extraction method 400 for extracting (i.e. de-watermarking) the authentication data embedded in the transmitted watermarked image at a receiver's end is depicted. In particular, the watermarked image is firstly divided into non-overlapping blocks in the same manner as what was previously carried out at the sender's end (i.e. see description pertaining to Figure 2). In this instance, it is therefore to be understood that the image is divided into a plurality of 2x2 pixel blocks. Each block is subsequently processed according to the following steps, listed in Figure 4, in a raster scan manner until the entire embedded bit stream is extracted.

At step S402, a first alterable pixel of a block, currently being processed, is selected. The first alterable pixel has a pixel value "A". Then at step S404, a value comparison is done between "A" and " ?', and "A" is then updated accordingly based on the result obtained. It is to be appreciated that the value of "FT used is the same as that previously used in the data embedding. The location of "R" is retrieved based on the encrypted location signal that was embedded in the watermarked image at known locations, in which the receiver already knows based on a prior arrangement made between the sender and receiver. In other words, the authentication data is embedded based on the randomly selected locations of "/?', while the encrypted location signal is embedded based on pre-fixed location of "FT. After retrieving the encrypted location signal, the encrypted signal is then decrypted using the private key of the receiver. The result is checked at step S406 in determining a next step to be taken in the flow of the data extraction method 400.

Specifically, if the result indicates that "A>R", then at step S408, the value of "A" is modified to "A-Q", and the updated "A" is consequently compared with "FT. It is to be appreciated that in this instance, "Q = Q ". At a next step S410, if the comparison determines that "\A-R\<Q", a binary bit value of "1 " is extracted; otherwise no bit is to be extracted.

On the other hand, if the result indicates that "A<R", the value of "A" is modified to "A+Q", and the updated "A" is compared with "R' at step S412. At a further step S414, if the comparison indicates that ")A-R}<Q", a binary bit value of "0" is extracted; otherwise no bit is to be extracted. On completion of either step S410 or step S414, the data extraction method 400 thereafter routes to step S416, in which steps S404 to S414 are to be repeated for other remaining alterable pixels in the current block. Once all the alterable pixels have being processed, data extraction for the current 2x2 pixel block is considered completed. The data extraction method 400 is then applied to a next 2x2 pixel block in the watermarked image , until all pixel blocks in the entire image have been processed to obtain the completed stream of authentication data. 3. Integrity Watermarking (as the Second Watermark Layer)

Integrity watermarking enables the detection and localization of image tamper, which may be introduced by a noisy transmission channel or a hacker, by embedding tamper detection watermarks on every zoned image region in a reversible manner. Specifically, the integrity watermarking method 1 04, based on the current embodiment, introduces a technique with high embedding capacity and fast processing, by incorporating a proposed method to exploit the 3-D characteristics of volumetric biomedical dataset for calculating CRCs and generate the two types of watermarks (i.e. horizontal and vertical watermarks). The integrity watermark is to be generated based on a volumetric image set of interest, in which the first image slice of the set is considered as the leading image and the other remaining image slices as non-leading images. In this respect, Figure 5 shows an embedding structure 500 in relation to the leading image 502, and a series of non-leading images 504. Particularly, the non- leading images 504 are embedded with only horizontal CRC, whereas the leading image 502 is embedded with both horizontal CRC and vertical CRC.

3. 1. Integrity Watermark Embedding for Non-leading Images

Correspondingly, Figure 6 shows a flow diagram of the method 1 06 for data embedding non-leading images 504, at the sender's end. Specifically, it is to be appreciated that the method 1 06 of Figure 6 relates to for the processing of a single non-leading image 504. At step S1 062, all pixel values of a non-leading image 504 to be processed are increased by an amount of "Q _? +Q , where "Q ₂ - 2Qi '. "Q ₂ " is a second watermarking embedding threshold used in the 3-D watermarking method 1 00 of Figure 1 . This is necessary in order to avoid overflow and underflow conditions before authenticity watermarking (while at the receiver's end, all the pixel values will also be decreased by "Q _? +0 ₂ " to restore the original image) . It is to be appreciated that the step of decreasing the pixel values by "Qj+Q performed at the receiver's end, when all the necessary data have been retrieved from both the first and second watermark layers. Thus, this step is performed as a last process of de-watermarking to restore the original image. Further, the second watermarking embedding threshold, "<¾", is defined to ensure sufficient embedding capacity for the second watermark layer.

In particular, "0 ₂ " maintains a comparable embedding capacity in the second watermark layer as the first watermark layer, because the first-layer watermark tends to double the difference between the reference pixel and alterable pixels in the worst case situation. The choice of both "Q " and "<¾" is also restricted by the underflow and overflow conditions, which occur in DICOM images with 1 6 bits depth when the gray level of a pixel, "n", is decreased by more than "n", or when the gray level of "65535-n" is increased by more than "n". However, medical imaging modalities generally do not produce images that utilize the full range of gray levels. As such, there is thus a need to increase all pixel values by "Or+O , as afore mentioned. At next step S1064, the non-leading image 504 is then divided into non- overlapping 1 8 x1 8 pixel blocks. Each 1 8 x1 8 pixel block is subsequently processed in a raster scan manner according to the following steps described hereinafter. At step S1066, the CCITT-CRC-16 bits of a currently being processed 1 8 x1 8 pixel block is computed. In particular, based on the current embodiment, a method of calculating the horizontal CRC to reduce the computation time required, as well as for maintaining tamper detection capability is performed in this step. Instead of using the 1 8 x 1 8 pixel block to calculate the horizontal CRC, a new 9 x 9 pixel block is generated from the current 1 8 x 8 pixel block by summing the pixel values of the 2 x 2 non-overlapping sub-blocks within the 1 8x 1 8 pixel block. This effectively reduces the input size required for the CRC calculation, and the CRC of the 9 x 9 pixel block is consequently more easily calculated.

At step S1 068, the data embedding method of Figure 3 is performed to embed the calculated CRC, by considering the current 1 8 x1 8 pixel block as an input image (i.e. W=H=1 8) . In other words, the calculated CRC is to be embedded with the current 18x1 8 pixel block. In addition, the segmented square block size is selected to be of size 3 x3 (i.e. N= 3). Furthermore, the central pixel is selected as the reference pixel for every 3 x3 pixel block, and in this instance, "Q ₂ " is defined as the watermarking embedding threshold "Q". The data embedding process then continues until all the CRC bits have been embedded.

Thereafter, a determination is made at step S 1 061 0 whether if all the 1 8 x1 8 pixel blocks of the current non-leading image 504 have been processed. If determination results in a "No" answer, the method 1 06 branches to the step S1 066 so that the steps S1 066 and S 068 are to be applied for the next 8 x 8 pixel block. Conversely, if the determination results in a " Yes" answer, the method 1 06 then branches to the step S1 061 2, where embedding of tamper detection watermark for the current non-leading image is considered successfully completed. It is also importantly to be understood that all the remaining non-leading images 504, included in the specific volumetric image set, are to be processed in a corresponding manner as described above. 3.2. Integrity Watermark Embedding for Leading Image

To speed up the tamper. detection on all the image slices, a proposed concept of vertical watermarking, to be utilised by the 3-D watermarking method 1 00 of Figure 1 of the current embodiment, is disclosed. Through use of vertical watermark, quick determination of whether a particular volume of 9 x9 pixel blocks (as shown in Figure 5) is tampered is accomplishable at the receiver's end. It is to be appreciated that each volume is defined as comprising a 9 x9 pixel block from the leading image 502 and a series of 9 x9 pixel blocks from the subsequent non-leading images 504 at the same horizontal coordinates (refer to Figure 5) . In the case that no tamper is detected, no further verification of the horizontal CRC is required for each image slice in the volumetric image set. In the case that tamper is detected in a volume of 9 x9 blocks, the horizontal CRCs of only the 1 8 x1 8 blocks which intersect with the detected volume of 9 x9 blocks need to be checked; the pair of failed vertical and horizontal CRC checks then reveal the exact location of the tampered region in the tampered image slice. In this way, tamper detection of the entire 3-D volumetric DICOM image set is accomplishable without necessitating exhaustive processing of all the image slices.

It is to be appreciated that the leading image 502 of the volumetric image set is selected to embed the vertical watermark, in addition to its own horizontal watermark. This may lead to more visual artifacts but after de-watermarking, both the horizontal and vertical watermarks are removed and the original image is then restored. Figure 7 is a flow diagram of the method 108 for data embedding the leading image 502, at the sender's end. Adopting similar data embedding technique as used for the non-leading images 504, all pixel values of the leading image 502 are increased by an amount of "Qj+Q ₃ " at step S1082, to avoid overflow and underflow conditions before authenticity watermarking. "Q ₃ " is a third watermarking embedding threshold used in the 3-D watermarking method 1 00, and "<¾" is defined to be "8Q," to ensure sufficient embedding capacity for the second watermark layer. Similarly, all pixel values are eventually to be decreased by "Qr+Q ₃ " to restore the original image at the receiver's end. It is to be appreciated that the step of decreasing the pixel values by "Q;+Q ₃ " performed at the receiver's end, when all the necessary data have been retrieved from both the first and second watermark layers. Thus, this step is performed as a last process of de-watermarking to restore the original image. Then the leading image 502 is divided into non-overlapping 1 8x 1 8 pixel blocks at step S1 084. Each 1 8 x 1 8 pixel block is subsequently processed in a raster scan manner according to the following steps described below.

It shall be appreciated that 'Ό ' is used for the first watermarking layer (for authenticity verification) while "(¾" and "Q ₃ " are used for the second watermarking layer (for tamper detection) . Hence, "Qj+Q ₃ " is therefore only applied to the leading image 502, while "Q _? +Q ₂ " is only applied to the non- leading images 504; in other words, the pixel values of the volumetric image are not effectively increased by a value of adding "Qj+Q to "Qj+Q ₃ ". It is also to be appreciated that "Q ", "<¾" and "Q ₃ " are primarily introduced to prevent the overflow and underflow problem.

At step S1086, the CCITT-CRC-16 bits of a currently being processed 18χ ί8 pixel block is computed using the same technique as described in step S1066 of the method 106 shown in Figure 6, which pertains to for processing the non- leading images 504. Next, at step S1088, the large 18x18 pixel block is divided into four non-overlapping 9 x9 pixel blocks, and in further step S1 0810, the respective vertical CRC-8 for each of the four volumes, where the four 9 x9 pixel blocks reside, are calculated. In particular, all 9 x 9 pixel blocks within a volume are (vector) summed to form one 9 x 9 pixel block, and the pixels in this resulting summed 9 x9 pixel block are then utilized to calculate the vertical CRC. Additionally, the data bits to be embedded in each 9 x9 pixel block are also subsequently assembled in this step, in which the entire stream of data bits comprises four data bits of the 1 6-bit horizontal CRC of the 1 8x 1 8 pixel block, as well as the 8-bit vertical CRC code of the volume where the 9 x9 pixel block resides. In other words, the length of the assembled data bits obtained for each 9x9 pixel block is configured to be 1 2-bits long.

In particular, it shall be appreciated that in respect of a single 1 8 x 1 8 pixel block, one 1 6-bit horizontal CRC is calculated. This 1 6-bit data is then divided into four partitions. Specifically, the first partition contains bits one to four, the second partition contains bits five to eight, the third partition contains bits nine to twelve, and the fourth partition contains bits thirteen to sixteen. Subsequently, each of the four 9 x 9 pixel blocks then embeds one corresponding partition of the afore described, in the following order, top left, top right, bottom left, bottom right. At step S 1 0812, the data embedding method of Figure 3 is performed on each 9 x 9 pixel block to embed the assembled data bits (being of length 1 2-bits) by considering the 9x9 pixel block as an input image (i.e. W=H=9) . The segmented square block size is then selected to be of size 3x3 (i.e. N= 3) . Moreover, the central pixel is selected as the reference pixel , and "Q ₃ " is used as the watermark embedding threshold "Q". A determination is made at step S1 0814 to ascertain if all the data bits have been embedded. If there are remaining data bits to be embedded, flow of the method 1 08 is redirected to the step S1 081 2, until all data bits are embedded for the current 9x 9 pixel block. A subsequent determination is then made at step S1 0816 to check if there any remaining unprocessed 9x9 pixel blocks. Thus, if there is any 9 x 9 pixel block that requires processing, the method 1 08 branches back to step S1 081 0, and from thereon, the necessary afore described steps are then performed with respect to the yet unprocessed 9x9 pixel block. In this manner, the data embedding process continues until all respective data bits (being of length 12- bits) have been embedded for each block. Conversely, if it is determined that all 9x 9 pixel blocks have been processed, the method 1 08 is concluded at step S1 081 8, which signals that embedding of tamper detection watermark for the leading image 502 is completed.

4. Tamper Detection and Localization

Figure 8 shows a flow diagram of a method 800 for detecting and localizing tampered areas of a watermarked image at the receiver's end, as part of a process for restoring the original image. At step S802 , the watermarked leading image 502 is first divided into non-overlapping 1 8x 1 8 pixel blocks. Each 1 8x 1 8 pixel block is then subsequently processed in a raster scan manner according to the following steps hereinafter. At next step S804, the 1 8x 1 8 pixel block, which is currently being processed, is divided into four non-overlapping 9 x 9 pixel blocks (ih the same way as done at the sender's end).

Then at step S806 , the data extraction method 400 of Figure 4 is performed on each 9x 9 pixel block which is considered as a watermarked image for the purpose at this stage. In particular, the 9x 9 pixel block is segmented into 3 x3 blocks (i.e. N= 3) to be similar as at the sender's end. It is to be appreciated that the central pixel is to be used as the reference pixel, and "O = Q ₃ ". The data extraction process continues until the respective 1 2-bits have been extracted for each 9 x9 pixel block. The respective extracted 1 2-bits data are individually being utilised to perform the necessary checking. Therefore, a total of 48-bits are to be extracted. Based on the extracted data bits, four extracted 8-bits vertical CRCs for the four volumes where the 9 x 9 pixel blocks reside, respectively, and one extracted 1 6-bit horizontal CRC for the leading image are obtained at step S808.

At step S81 0, a determination is made to check whether the leading image 502 has been tampered based op the extracted data bits. More specifically, for each 9x 9 pixel block, if the number of bits extracted is less than "1 2" or the extracted 8-bit vertical CRC does not match with the newly calculated vertical CRC, the volume where that 9 x9 pixel block resides is then identified as being tampered. The extracted 6-bit horizontal CRC is subsequently used to check whether the 1 8 x 1 8 pixel block of the leading image is tampered. In particular, once a vertical CRC indicates that tampering has occurred, the image slices (comprising both leading and non-leading images 502, 504) within the volumetric image that have been tampered are to be determined. In order to achieve that, the 16-bit horizontal CRC needs to be checked, and is extracted as afore described. If the extracted 16-bit horizontal CRC does not match with the newly calculated horizontal 16-bit CRC, then tampering is ascertained to have occurred.

On the above discussed basis, a tampering detected decision is taken at step S812. If tampering is detected, the method 800 branches to step S814, where further examination on the 18x 18 pixel blocks of each non-leading image in the same volume as the 18x 18 pixel block of the leading image is required to determine the exact location of the tamper. This examination process will be discussed shortly with reference to Figure 9. Conversely, if no tampering is detected, the method 800 directs to step S816, where each of the four volumes of 9x9 pixel blocks is identified as tamper free. The 18x 18 pixel block of each non-leading image that resides in the same larger volume as the 18x 18 pixel block of the leading image is then processed only using the data extraction method 400 of Figure 4, until all 16-bits have been extracted in step S818. Particularly, the 16-bits CRC are extracted to recover the original image (reversing the watermarking process), and further the 1 6-bits CRC extracted in this step S818 is the same as the 1 6-bits CRC obtained at step S808. It is to be appreciated that no further verification of the extracted 16-bits CRC is required at this stage of the processing. Referring to Figure 9, which depicts a flow diagram of a method S814 for further tampering examination as used in the step S814 of Figure 8, the examination process is carried out as follows. At step S8142, the data extraction method 400 of Figure 4 is performed on the current 1 8x 18 pixel block, which is considered as a watermarked image in this instance. The 18x 1 8 pixel block is also segmented into a plurality of 3x3 blocks, to be similar as at the sender's end. At further step S8144, the 16-bits CRC is extracted, which are considered to be the extracted CRC code, or until all the available bits have been extracted. Then at next step S8 46, it is to be determined either if the extracted CRC matches with the calculated CRC, or if the extracted bit length is sixteen. A determination decision, based on the above criteria, is next taken at step S8148, and if the number of bits extracted is however less than sixteen, or the extracted CRC code does not agree with the newly calculated CRC code of the restored image block (i.e. does not agree with the criteria), the block from the tampered volume of 9x9 pixel blocks identified earlier that intersects with the 18x 18 pixel block is identified as being tampered in step S81412. This then concludes the tampering examination process at step S81412. On the other hand, if the determination decision is found to be in agreement with the defined criteria, then no tampering is detected in the current non-leading image 504 as determined at step S81414. The 18x 18 pixel block of a next non-leading image 504 (if there are any remaining ones) is then processed, in which the method S814 is redirected back to step S8142 to restart the entire examination process (i.e. steps S8142 to S8148) for the next non-leading image 504, until there are no unprocessed non-leading images 504 left in the volumetric image set, or otherwise, tampering is identified in one of the subsequent non-leading images 504 being examined.

Therefore, in using the method 800 of Figure 8 (and in conjunction with the method S814 in Figure 9), either tampered areas are detected and localized, or the watermarked image subsequently proceeds to the extraction stage of authenticity watermarking. Consequently, the original image is then restored.

It is to be appreciated that the sequence of de-watermarking the watermarked image is determined based on the prior sequence of watermarking performed on the image at the sender's end. In this embodiment, during the watermarking stage, authenticity watermarking is first performed, and followed by integrity watermarking. As a result, during the de-watermarking stage, integrity de- watermarking is to be first performed, followed by authenticity de-watermarking. In other words, de-watermarking follows the reverse sequence of watermarking. Moreover, performing tampering detection of the watermarked image before carrying out authenticity de-watermarking allows for checking of the integrity of the authenticity data that were embedded into the image.

^■ 5. Performance Evaluation

The 3-D watermarking method 100 of Figure 1 is evaluated in terms of its computational efficiency and tamper detection capability using four test sets of 3-D volumetric DICOM images with the parameters summarized in a first table 1000 shown in Figure 10. The test images are MRI images of different anatomical structures. The evaluation was performed by running MATLAB (from The - Mathworks Inc., Natick, MA, USA) program on a laptop configured with a 1 .83GHz dual-core CPU processor and 2GB of RAM. Higher computational efficiency is a vital factor of a practical solution, hence both the watermarking and de-watermarking times are compared between the proposed 3-D watermarking method 100 and a prior art method, repeated over the test volumetric image sets. Tamper detection capability is evaluated by performing both systematic tampering and real-case tampering. Systematic tampering is used to examine the tamper localization accuracy and tamper display resolution, whereas real-case tampering is used to assess the tamper detection and localization function from a medical practice point of view.

6. Results of Evaluation

In using the proposed 3-D watermarking method 100, Figure 1 1 (a) shows screenshots of evaluation results corresponding to successful watermarking of a MRI heart dataset, while Figure 1 1 (b) are corresponding screenshots of successful de-watermarking of the same dataset. The watermarked images displayed under the watermarked image files in Figure 1 1 (a) are the same images as the watermarked images displayed under selected watermarked image files in Figure 1 1 (b). It is to be highlighted that this set of evaluation results are obtained under when no tampering was effected on the MRI heart dataset. Further, the 3-D watermarking method 1 00 is also capable of detecting and localizing tamper performed on the test images, which is illustrated in Figure 12. In particular, Figure 12 shows screenshots 12 evaluation results corresponding to successful tamper detection and localization, ^" with the identified tampered image slices "1 ", "2", "5", and "8" shown enlarged, and the tampered areas are indicated in red square boxes. It is to be noted that the images were tampered by altering the pixel values of the watermarked images at various locations of the heart in the MRI images.

. ^' 6. 1. Computational Efficiency

The four test sets (of volumetric images) listed in the table in Figure 10 were used to evaluate computational efficiency of the proposed 3-D watermarking method 1 00 under tamper-free circumstance. The processing time taken to watermark each image from the test sets using a prior art method, and the 3-D watermarking method 1 00 was recorded and the respective average time required for each set is listed in a second table 1300 depicted in Figure 13. From the second table 1300, it is observed from the results that the 3-D watermarking method 100, on average, achieved a significant 65% reduction in watermarking time. This shows that the 3-D watermarking method 100 significantly improves the computational efficiency of watermarking 3-D volumetric images.

Further, the time to process each image from the test sets during the de- watermarking stage using the prior art method, and the 3-D watermarking method 100 was recorded. The respective average time required for each set is then listed in a third table 1400 shown in Figure 14. Results recorded in the third table 1400 show that the 3-D watermarking method 100 achieved a considerable reduction in de-watermarking time, an average of 72% reduction. The savings gained in de-watermarking time is significant because when there is no tampering, only the vertical CRC checks are performed, and no horizontal CRC checks are necessary.

The foregoing results therefore demonstrate that the proposed 3-D watermarking method 1 00 is able to effectively improve the computational efficiency by reducing the required input image size used for CRC calculation, and further, in using the vertical watermarking technique, makes the 3-D watermarking method 100 a better solution, over existing solutions, for usage in practical situations. 6.2. Tamper Detection Capability

6.2. 1. Systematic Tampering

Systematic tampering of watermarked images is performed by tampering a single pixel at random locations in randomly selected images in a volumetric set that has been watermarked by the 3-D watermarking method 100. Four images are then randomly selected, and each image is tampered at two random pixel locations using MATLAB. It is found that the 3-D watermarking method 100 achieved 100% tamper localization accuracy for all the systematic tampering tests. Sample evaluation results of successful tamper localization for the heart image set are shown in the screenshots 1500 of Figure 15. In particular, image slices "2" and "5" of the MRI heart image set detected with the tamper have been localized using the 3-D watermarking method 1 00. Red square boxes in the image indicate where the tamper is detected. It is also to be highlighted that the tampered image slices may be mistaken as un-tampered without aid of the 3-D watermarking method 00 for assessment in this respect.

Furthermore, compared with the prior art method, the 3-D watermarking method 100 uses a much smaller area to display the tamper localization result (as shown in the screenshots 1 600 of Figures 16a and 16b). Figure 16a specifically shows that image slice "1 " of the MRI heart image set with tamper has been localized using the prior art method, and each red square box has a size of 1 8 x1 8 pixels. On the other hand, Figure 16b shows that the image slice "1 " of the MRI heart image set with tamper has been localized using the 3-D watermarking method 100, and each red square box, in comparison to those for the prior art method, has a smaller size of 9 x9 pixels. The smaller display area is achieved by using the vertical CRC to locate the tampered volume of 9 x9 pixel blocks (refer to Figure 5), and then using the horizontal CRC to locate the exact 1 8 x1 8 pixel block that is tampered.

The intersection of the tampered volume of 9 x9 pixel blocks with the located tampered 1 8x 1 8 pixel block is thereafter determined to be the tampered area. Consequently, as the determined intersection is indicated by a small 9 x9 pixel block, a better resolution for examining the tampered areas is achieved in using the 3-D watermarking method 00 for this purpose. 6.2.2. Real-Case Tampering

Real-case tampering of watermarked images refers to tampering of image regions that are of clinical importance, with an intention to deceive the doctor. A set of knee MR images was used to evaluate the proposed 3-D watermarking method 100, in which the tampered images are simulated as being maliciously altered by such professional tampering attempts. For this set of evaluation study, three consecutive image slices from the knee set were altered by thinning the femoral cartilage to signify cartilage damage.

Based on the results, the 3-D watermarking method 100 is found to be able to detect and localize all the tampered regions in the test set. The sample results of the knee MR image set are displayed in the screenshots 1 700 of Figures 17a to 17c. Particularly, Figure 17a shows the watermarked image slices "13" and "14" of the knee MR image set without tampering, whereas Figure 17b shows the watermarked image slices "13" and "14" of the same set with tampering, and finally Figure 17c shows the watermarked image slices "13" and "14" with tampered locations, correctly identified (as indicated by red square boxes). These results demonstrate that the 3-D watermarking method 100 of tamper detection and localization are competent for real life scenarios, in which multiple pixels are tampered to realistically deceive an intended reader (e.g. the doctor).

Furthermore, compared with the result of tamper localization using the prior art method, the 3-D watermarking method 1 00 shows a feature of being able to restrict the detected tampered area into a smaller localization area of 9 x9 pixel blocks, as shown in the screenshots 1 800 of Figure 1 8. More specifically, Figure 18a is the image slice "1 " of the knee MR image set with tamper being localized using the prior art method, and each red square box has a defined size of 1 8 x 1 8 pixels, while Figure 1 8b shows the same image slice "14" of the knee MR image set with tamper now being localized using the 3-D watermarking method 1 00, and each red square box has a smaller size of 9 x9 pixels.

7. Discussions

The fully reversible digital 3-D watermarking method 100, as depicted in Figure 1 , for the protection of 3-D volumetric DICOM images importantly improves the computational efficiency of the watermarking algorithm. The overall improvement makes the 3-D watermarking method 1 00 more viable to be used in a practical medical environment. The reversibility of the watermarking technique also allows the original medical images to be restored for medical practice, which has been considered by radiologists as an important clinical requirement. Additionally, use of public key cryptography prevents unauthorized personnel from retrieving the embedded authentication information to gain access to the original image data. Consequently, this ensures the confidentiality and authenticity of the DICOM medical images. Another important feature of the 3-D watermarking method 100 is facilitation of tamper detection and localization capabilities. This feature is achieved mainly by checking whether if there are any variations with the block-based CRC.

For the proposed 3-D watermarking method 100, improvements in embedding capacity and processing time are advantageously achieved through a number of characterising features. Firstly, a 3 x3 pixel block is selected as the basic embedding structure instead of using a 2 x2 pixel block, according to the prior art, in order to increase the embedding capacity. By using a 3 x3 pixel block, the maximum number of the alterable pixels within a block is significantly increased from three pixels to eight pixels too. Because alterable _. pixels are the sources of embedding capacity according to the embedding algorithm, embedding capacity is accordingly increased.

Secondly, the area for embedding each horizontal CRC is chosen as 1 8 x1 8 pixels, instead of 1 6 x 1 6 pixels. This change was adopted so that the 1 8 x 1 8 pixel block may fully be filled with the newly defined embedding blocks of 3 x3 pixels, and so that there are enough embedding capacity for horizontal CRC.

Thirdly, a larger watermark embedding threshold "Q ₂ " is defined to ensure that each 1 8 x1 8 block has enough embedding capacity for the horizontal CRC, regardless of the bit depth of the image to be watermarked. All these improvements enable each embedded horizontal CRC to be confined strictly within its own 1 8 x 1 8 pixel block; otherwise certain CRC may have to be embedded across other blocks due to insufficient alterable pixels. This has a beneficial effect of allowing tamper detection to be more localized and prevents tamper localization from propagating to a few blocks.

Fourthly and more importantly, the method proposed at step S1 066 of Figure 6 is able to reduce the horizontal CRC computation time. This method leads to a significant reduction in the watermarking time because typically, majority of the processing time is contributed by the horizontal CRC calculation. This significant improvement in computational efficiency desirably makes the proposed 3-D watermarking method 1 00 a more practical solution for protecting 3-D volumetric DICOM images, in terms of ensuring the end integrity of the transmitted images.

It is to be highlighted that the 3-D watermarking method 100 further introduces the concept of vertical watermark, which exploits the characteristics of multiple image slices contained within a volumetric image dataset to reduce processing time incurred during de-watermarking, and to improve the tamper localization resolution. Moreover, using the vertical watermark brings benefits in three aspects, albeit at the expense of incurring a small increment in the watermarking time (which is however considered trivial compared with the time savings resulting from the proposed method of calculating horizontal CRC and the proposed de-watermarking technique for tamper detection).

With regard to those three aspects, the use of vertical watermark firstly decreases the de-watermarking time by avoiding the need to perform checks on multiple horizontal CRC for non-leading images in tamper-free situations. Secondly, the 3-D watermarking method 100 may even detect whether if there are missing image slices with the help of vertical watermark, a capability that is not available and unachievable using existing methods. Lastly, when image tamper is detected, the 3-D watermarking method 100 is able to provide a smaller area (and hence better spatial resolution) for tamper localization display.

It is to be appreciated that although MATLAB programming language was used for the purpose of evaluating the 3-D watermarking method 1 00, C++ programming may however also be used to implement real-time software to further speed up the processing time. Moreover, the power of parallel computing may be utilized in the future, which allows further improvement of computational efficiency. In summary, the efficient 3-D watermarking method 100 with improved tamper detection and localization capability for volumetric DICOM images is disclosed. The 3-D watermarking method 100 utilises watermark generation and detection algorithms that are developed to significantly reduce the computational processing time and improve the tamper localization capability without compromising authentication accuracy (which is provided using public-key encryption techniques) . Specifically, the 3-D watermarking method 100 involves using two different types of watermarks: horizontal and vertical watermarks, in which the former parses the three-dimensional volumetric data along the image slices, and the latter, across the image slices. Furthermore, an adaptive data insertion algorithm is also disclosed and used for the insertion of data in order to ensure that there is sufficient embedding capacity regardless of the bit depth of the image. The performance of the proposed 3-D watermarking method 100 was evaluated by using real volumetric DICO images. Results show that the 3-D watermarking method 100 achieved approximately 65% and 72% reduction in watermarking and de-watermarking processing time respectively (compared to straightforward image-by-image processing), and was able to ensure authenticity and integrity of ^' the images. Further, in cases where the images were tampered, the 3-D watermarking method 100 is also able to detect and localize the tampered regions with better resolution (compared to existing methods which are typically developed for two-dimensional, instead of volumetric, DICOM images).

The described embodiment(s) should not however be construed as limitative. For example, it is to be appreciated that the 3-D watermarking method 1 00 may employ only the authenticity watermarking method 102 to create the first watermark layer in a volumetric image, but not include the second watermark layer (i.e. does not use the integrity watermarking method 104 in this instance). Alternatively, the 3-D watermarking method 100 may also be configured to be able perform the reverse, i.e. only create the second watermark layer in the volumetric image, but not include the first watermark layer (and thus does not use the authenticity watermarking method 1 02 in this regard) . In other words, the first and second watermark layers may each independently be generated and embedded in the volumetric image, without requiring use of the other.

Moreover, the proposed 3-D watermarking method 100 may also be used for watermarking and de-watermarking volumetric images that are not necessarily medical images, but also other types of volumetric images, as such for example, 3-D volumetric images generated for geological exploration purposes or related to confocal imaging data. It is also to be appreciated that the leading image 502 may be any selected image slice in the volumetric image set, while the non-leading images 504 are correspondingly the remaining image slices (which exclude the selected image slice used as the leading image 502) ; in other words, the leading image 502 may be located in between the non-leading images 504. Further in this instance, the receiver will need to know which image slice is designated and used as the leading image 502.

In the described embodiment, CRC is used as error detection codes but other error detection codes are envisaged.

While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary, and not restrictive; the invention is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practising the claimed invention.