连接大米籽粒图像的自动分割算法研究

Automated Identification and Separation of Touching Rice Grains with Machine Vision

分析了单个大米籽粒和连接大米籽粒的轮廓特征,提出了一种基于曲率的分割算法。该算法能够有效地反映边界点尖锐程度的曲率及其方向,根据曲率大小和曲率方向能准确判别图像中大米籽粒是否接触并快速找到分割点。用最短距离配对法对3种接触的大米籽粒进行分割。算法对判别籽粒是否接触的准确率达99%以上,对接触籽粒的准确分割率达95%以上。

Automated identification and separation of touching kernels have important effects on result of quality detection of rice kernel with machine vision. This paper analyzed profile shape of isolated rice kernel and touching kernels and developed an algorithm based on the curve. The algorithm determined the curve and its direction which can reflect sharp degree of boundary. It can find quickly whether rice kernels touched or not and separated points. The method of minimum distance was used to segment touching kernels. For three types of rice kernels, algorithm could perform over 99% accuracy for identifying touching or not and can perform over 95% for segmenting.