基于机器视觉的粘连米粒图像分割算法研究  

Research on Image Segmentation of Touching Rice Based on Machine Vision

作　　者：王粤[1] 黄肖琳 余杭州 葛光涛[1] Wang Yue;Huang Xiaolin;Yu Hangzhou;Ge Guangtao(Zhejiang Gongshang University, Hangzhou 310018)

机构地区：[1]浙江工商大学,杭州310018

出　　处：《中国粮油学报》2018年第6期106-112,共7页Journal of the Chinese Cereals and Oils Association

基　　金：浙江省自然科学基金(LY14F010001)

摘　　要：提出了一种新的基于机器视觉的大米粘连米粒图像的分割算法。由于在利用机器视觉方法检测抽样米粒的各种外观指标时,米粒是随意放置在载样台或传送带上的,有些米粒会不可避免地互相碰触,由此造成在图像预处理时获取的米粒轮廓线互相粘连在一起,从而在后续的检测中被误判为是一颗米粒,致使大米外观指标如碎米率、垩白米率、黄米粒率的检测都出现错误。本研究根据粘连米粒的内轮廓及外轮廓线上的粘连点的特征,提出了一种新的能快速精确检测米粒的粘连点的算法,通过计算各粘连点之间的距离及曲率方向,准确地判断出互相配对的最佳粘连点对,从而实现了粘连米粒的精确分割,试验结果表明该方法的准确性和高效性。This paper proposed an image segmentation algorithm of touching rice based on machine vision.When we use machine vision to detect the appearance quality index of rice,we often put rice sample on the objective table,the rice could touch together,then when we get the contour of rice images,the touching together rice’s contour will be thought as one rice,then the successive detection of ration of broken rice,yellow rice and chalk rice will be wrong. In this paper,according to the feature of touching points on the outer and inner rice contour,we proposed a fast and high precise algorithm to detect all touching points on the outer and inner rice contour,and then the best pairs of touching points can be determined precisely based on the distant and curvature direction of touching points.Experiment result indicated that our algorithm achieved significant improvement.

关 键 词：图像分割 粘连 曲率 机器视觉 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]