An algorithm was proposed to overcome the difficulty that defects could not be successfully segmented at one time due to illumination nonuniformity on fruit surface. Navel orange was selected as experiment object. Firstly, the R component image of navel orange was extracted and original image was obtained by removed R component image background. Then, incident component of original image was obtained by low pass filtering based on illuminationreflectance model. The incident component was considered as illumination image of original image. Finally, the corrected image was computed by original image dividing by illumination image. Defects were successfully segmented at one time by a threshold value. The sound samples and samples with other ten types of peel diseases were detected by using the proposed method. The experimental result showed that the recognition rate was over 97% based on 788 obtained images.