Automatically Determining the Segmentation Lines between Images of Adherent Rice Grains

Abstract

Abstract. A new strategy for accurately and automatically determining the segmentation lines between images of adherent rice grains was explored based on the characteristics of contour and the techniques of image analysis. The images were captured by using an optical scanner. The original images of touching rice particles were preprocessed by the techniques of image analysis including de-noise, segmentation, and contour extraction. The extracted contours were smoothed by convoluting with Gaussian kernel. The first-derivative of the smoothed vectors was computed to look for the local peaks which corresponded to the segmentation nodes. A radiate triangle region was proposed to determine the confidence region for the potential matching nodes. The direction of radiate region was determined in terms of the axis tilted from the normal through the curvature center. The nodes which fell into the triangle region might become the candidate matching nodes. The splitting line linking between the matching nodes with the shortest distance was considered as the correct splitting path. The results demonstrated that the proposed method could be successfully used to split the images of touching rice particles. Keywords: Confidence region, Image analysis, Matching, Rice particles, Segmentation.