统计区域合并的彩色图像分割算法

西安交通大学 电子与信息工程学院,陕西 西安 710049

彩色图像分割; 观测图像; 统计区域; 合并预测

Color image segmentation method of statistical region merging
GUO Xin

(School of Electronic and Information Engineering,Xi'an Jiaotong University,Xi'an 710049,China)

color image segmentation; observed image; statistical region; merging prediction

DOI: 10.13800/j.cnki.xakjdxxb.2015.0320

备注

针对传统区域合并算法中存在的分割复杂度高、分割精度低的问题,提出一种将统计理论应用于区域合并的彩色图像分割算法,该算法通过建立图像生成模型,得到新的合并预测准则,有效地避免合并过程中可能导致的区域边界破坏,提高分割精度,降低分割复杂度。在对已有算法分析的基础上,提出基于古典概率理论的图像生成模型,重点介绍区域合并思想与统计理论相结合的合并预测准则,该准则是逐步松弛的,确保在无像素遗漏的同时分割的精度。算法不但考虑了像素的相似性,还考虑了空间上的邻接性,因此可以有效消除孤立噪声的干扰。通过与基于连接图的系统工程分割方法比较发现,文中算法的运算时间具有明显优势。实验结果表明,该算法还具有较高的分割精确度和较强的鲁棒性,分割尺度可调。

Aiming at the traditional region merging segmentation algorithm with the problems of high complexity and low accuracy,an image segmentation method based on statistical region merging is proposed.Through the establishment of image generation model,this paper obtains the new merging prediction criteria which effectively avoids the region boundary damage in the merging process,improves the segmentation accuracy and reduces the segmentation complexity.Based on the analysis of existing algorithm,a model of image generation is generated using classical probability theory,then the merging prediction criteria as a blend of the region merging and the statistical property is explained in detail.This criteria is gradually relaxed,ensuring segmentation accuracy without pixel omission.This algorithm not only considers the similarity of pixel but also the adjacency space,therefore it can effectively eliminate the interference of isolated noise.By comparison with the Joint Systems Engineering Group Segmentation Method,this algorithm has obvious advantages in computation time.Experimental results show that the algorithm with the adjustable segmentation scale is of high segmentation accuracy and strong robustness.