A new method to detect surface defects of fruits

Segmentation of gray level image is not easy when the edge of the object is not clear enough. The Convexity of Illumination Distribution (CID) feature can be utilized to recognize objects from the background in this situation. In this paper, we propose three models to prove the convexity of the illumination distribution. Therefore, we are able to find support points using convex hull and then reconstruct the background from these points. Finally, we subtract this background from the original picture and set a threshold to make segmentation and get object. In order to test the efficiency of this methodology, quantitative experiments are performed and yield promising results. We also compare the segmentation results between different methods to show the efficiency of this method.