An original image is read as an aggregate of a plurality of pixels in which adjacent pixels have different colors (R, G, and B) in a main scanning direction and in a sub-scanning direction, and the read pixels of the respective colors are stored in a line memory in association with information on relative positions of the pixels with respect to another pixel. Then, the stored pixels are sorted so that pixels having the same color are adjacent to each other, and an abnormal pixel (dust) not present in the original image is detected based on the state of the sorted pixels. With this, the dust not present in the original image is detected without increasing the cost, and the dust is corrected without forming a conspicuous trace of correction.