Shadow removal method of apples based on illumination invariant image

Abstract: Rapid and accurate recognition of apple target with shadows on its surface is one of the key problems which must be solved for apple picking robot's vision system. In order to realize rapid and accurate recognition of apple target under influence of shadow, a shadow removal method based on illumination invariant image was proposed. Firstly, the red component image of original image was extracted, which can highlight the unshaded area and high brightness area of apple, and keep the shadow areas; Secondly, the illumination invariant image of original apple image was extracted. The illumination invariant image obtained highlights the shadow areas and weakens the areas of strong light, which is just opposite to red component image. Thirdly, the apple image with shadow removal could be obtained by adding the illumination invariant image to red component image, which could eliminate the shadow areas effectively. Finally, Adaptive threshold segmentation algorithm was adopted to detect the apple target from the image with shadow removal. In order to verify the validity and the accuracy of the proposed method, 20 apple images affected by shadow which were captured in the natural scene were tested. The performance of the proposed method was compared to that of Otsu method and chromatic aberration segmentation algorithm based on 1.5*R-G. The result showed that the segmentation result of Otsu algorithm was very poor which could only identify the unshadow areas of apple and could not identify the shadow areas; chromatic aberration segmentation algorithm based on 1.5*R-G was greatly influenced by light, which could not identify strong light areas and some shadow areas of image; while the result of shadow removal method of apples based on illumination invariant image was better than these two methods. The proposed method can not only identify apples affected by shadow area which was caused by illumination, but also overcome the influence of the strong illumination. The average FPR of proposed method was 17.49%, which was decreased by 52.84% and 26.18% respectively, compared to Otsu algorithm and chromatic aberration algorithm based on 1.5*R-G. The average OI was 86.59%, which was increased by 47.2% and 11.03%, compared to Otsu algorithm and chromatic aberration algorithm based on 1.5*R-G. Thus, it could be concluded that apple images under influence of shadow can be effectively identified by the proposed method in this paper, which is feasible in identifying the apple target with shadow on its surface.