最小类内方差和区域生长相结合的图像分割法

Image segmentation algorithm combining minimum interclass variance with region growing

针对视频车辆检测中,光照不均匀、对比度不强的多目标图像分割,提出了一种基于最小类内方差和区域生长相结合的快速阈值分割算法.首先对最小类内方差法进行改进,快速确定差分图像的最佳分割阈值,再用区域生长法分割得到目标.理论分析和实验结果表明,该分割算法不仅适用于简单的车辆图像分割,而且对于复杂的车辆图像也取得了较好的分割效果.该算法计算量小,分割精度有一定优势,有助于下一步的目标识别.

A kind of fast threshold segmentation algorithm based on the minimum interclass variance and region growing was proposed for the multi-target image segmentation with uneven light and infirm contrast ratio in the video inspection of vehicles. Firstly, the minimum interclass variance method was improved to obtain the optimal segmentation threshold of difference image quickly, and then, the region growing method was used to segment the target. The theoretic analysis and experiments indicate that the present segmentation algorithm can not only fit for the segmentation of simple image, but also be suitable for those complex images. Moreover, the present segmentation algorithm has the higher segmentation precision and lower computation complexity, which is helpful for further target recognition.