Application of Zernike-moment-based watershed segmentation on fruit features extraction

Abstract: Image segmentation and the extraction of target contour are key technologies to realize continuous non-destructive detection of crop geometric features based on machine vision, which can help understanding the development of internal physiological and ecological changes in fruits and other crops during growth. In this paper, a new watershed algorithm based on Zernike-moment edge detection was proposed to extract grape fruit contour features. The algorithm first acquired a rough outline of the target based on watershed algorithm marked by Zernike-moment. Then through contour refinement template algorithm, a single-pixel contour was obtained. In the meanwhile, false edges were removed by template algorithm. Finally, contour tracking was conducted to get the real contour of the target. Compared with traditional marker driven watershed algorithm, the proposed algorithm can avoid contour destruction caused by tag via Zernike-moment edge detection. In this way, the target contour is well preserved, so that post-processing can be reduced and detection efficiency is improved. At last, by comparing the contour obtained by the proposed algorithm with that by traditional marker driven watershed algorithm, the feasibility of the algorithm was demonstrated. Because of high performance of detection, the algorithm is able to meet the requirements of continuous contour feature extraction of grape fruit. The method can be applied to the real-time detection of grape fruit geometrical feature changes.