Liu Hongquan, Zhang Weijin, Wang Fushun, Sun Xiaohua, Wang Junhao, Wang Chen, Wang Xinxin
College of Urban and Rural Construction, Hebei Agricultural University, Baoding, 071000, China.
College of Information Science and Technology, Hebei Agricultural University, Baoding, 071000, China.
Heliyon. 2023 Apr 15;9(4):e15097. doi: 10.1016/j.heliyon.2023.e15097. eCollection 2023 Apr.
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 ( 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.
作为图像处理中的一个重要步骤，图像分割可用于确定物体计数、面积和轮廓数据的准确性。此外，图像分割在种子检测研究中不可或缺。由于原始图像的灰度级不均匀，传统的分水岭算法会产生许多错误的边缘，导致过分割和欠分割，从而影响获取种子表型信息的准确性。本文提出了一种基于距离图重建的改进分水岭算法——DMR-分水岭算法。根据图像的灰度分布特征，选择灰度缩减幅度h生成与原始图像灰度分布趋势相同的掩膜图像。根据待分割的不同区域的虚假最小值选择相应的阈值对原始灰度图进行重建，生成一个消除错误边缘的精确距离图。选取小豆（Vigna angularis (Willd.) Ohwi & Ohashi）图像作为实验材料，在两粒子粘连和多粒子粘连两种情况下研究各算法分割计数结果的残留率。将所提算法的结果与常用于种子分割的传统分水岭算法、边缘检测算法和凹点分析算法的结果进行比较。在两粒子粘连的情况下，分水岭算法和边缘检测算法的残留率分别为0.233和0.275，而凹点分析算法的残留率为0，证明其适用于两粒子粘连。在多粒子粘连的情况下，凹点分析算法不适用，因为它会破坏种子图像。分水岭算法和边缘检测算法的残留率分别为0.063和0.188，而所提算法在两粒子粘连情况下的残留率为0，计数准确率达到100%，证明了所提算法的有效性。本文算法显著提高了粘连种子图像分割的准确性，为种子检测中的图像分割处理提供了新的参考。
