To reduce the extraction of erroneous identification area images in a landscape image by collating the candidate for an identification area image extracted in a visible light image with an infrared image and inspecting the contour pixels of the candidate of the identification area image.
A visible light image of a landscape photographed by a visible light camera 1 is inputted to an identification area image extracting means 3, and a candidate R for an identification area image is selected by using the characteristics of an identification. Meanwhile an object contour extracting means 4 extracts the contour of an object from an infrared image indicating the temperature of the visible light image photographed with an infrared camera 2 to generate a contour image Gi'. An identification area image candidate evaluating means 5 calculates the complexity Fv of the candidate R and the complexity Fi of an area corresponding to the candidate R in the image Gi', respectively determines threshold for the complexities Fv and Fi and narrows down the candidate R. Next, the number of pixels of a contour existing in an outer circumferential neighborhood area of the candidate R in the image Gi' and the ratio Fp of contour length of the candidate R are found. An identification area image is extracted in the candidate R by using the ratio Fp.
MIYAMOTO NOBUO
MACHIDA MOTOHIRO
NAKAMURA OSAMU
SUGIMURA TOSHIAKI