Application of an Improved Watershed Algorithm Based on Distance Map Reconstruction in Bean Image Segmentation
Hongquan Liu,Weijin Zhang,Fushun Wang,Xiaohua Sun,Junhao Wang,Chen Wang,Xinxin Wang
DOI: https://doi.org/10.1016/j.heliyon.2023.e15097
2022-01-01
SSRN Electronic Journal
Abstract:As an important step in image processing, image segmentation can be used to determine the accuracy of object counts, and area and contour data. In addition, image segmentation is indispensable in seed testing research. Due to the uneven grey level of the original image, traditional watershed algorithms generate many incorrect edges, resulting in oversegmentation and undersegmentation, which affects the accuracy of obtaining seed phenotype information. The DMR-watershed algorithm, an improved watershed algorithm based on distance map reconstruction, is proposed in this paper. According to the grey distribution characteristics of the image, the grey reduction amplitude h was selected to generate the mask image with the same grey distribution trend as that of the original image. The original greyscale map was reconstructed with corresponding thresholds selected according to the false minima of different regions that are to be segmented, which generates an accurate distance map that eliminates the wrong edges. An adzuki bean (Vigna angularis L.) image was selected as the experimental material and the residual rate of the segmentation counting results of each algorithm was investigated in two cases of two-particle adhesion and multiparticle adhesion. The results of the proposed algorithm were compared with those of the traditional watershed algorithm, edge detection algorithm and concave point analysis algorithm which are commonly used for seed segmentation. In the case of two-particle adhesion, the residual rates of the watershed algorithm and edge detection algorithm were 0.233 and 0.275, respectively, while the residual rate of the concave point analysis algorithm was 0 which proved to be suitable for two-particle adhesion. In the case of multiparticle adhesion, the concave point analysis algorithm was not applicable because it would destroy the seed image. The residual rates of the watershed algorithm and edge detection algorithm were 0.063 and 0.188, respectively, while the residual rate of the proposed algorithm in the two-particle adhesion cases was 0 and the counting accuracy reached 100%, which proved the effectiveness of the proposed algorithm. The algorithm in this paper significantly improves the accuracy of image segmentation of adherent seeds, and provides a new reference for image segmentation processing in seed testing.