| abstract |
(57) [Summary] (with correction) [Problem] To remove abnormalities in digital images such as dust and scratch images from images, false positives, that is, the number of non-abnormal image parts that are erroneously recognized as abnormal. To minimize SOLUTION: Initial recognition of an abnormal image area which passes first is performed based on a difference between an image gradient of each grid point of an image and an average image gradient of neighboring points. If the change in the pixel value of the most recent original image is greater than the change in the neighboring pixel values in a wider range, the area is recognized as abnormal. This technique significantly reduces the number of false positives detected in noisy image areas, such as daytime leaves and pebble beaches. After the list of the abnormal candidate areas that passed first is created, the list is thinned out by a series of detection indices. This set of detection indices can include shape, size, color, and visibility indices. These indices indicate how close the abnormal candidate is to dust particles and scratches. |