Abstract: Infrared human images always have low contrast, blurry boundaries and poor ability to distinguish details, and requirement of real time processing. To overcome above problem, a novel and effective method is proposed to infrared human images segmentation. The generalized fuzzy entropy is constructed based on Renyi entropy principle, namely fuzzy Renyi entropy. For getting threshold in shorter time, an improved simulated annealing algorithm based on chaos search is designed to search optimal threshold. Then, the proposed fuzzy entropy is utilized in infrared human images segmentation combined with chaos simulated annealing algorithm, and the performance is compared with several famous image threshold segmentation method. Experimental results demonstrate the approving segmented results are obtained, the robustness is better than other method, and the CPU time is 0.8s to 256 level grayscale images when the presented method is used in infrared human images segmentation. It can satisfy the requirements of accuracy, real time of infrared images processing.

Key words: infrared human image segmentation, fuzzy Renyi entropy, chaos simulated annealing

摘要： 针对红外人体图像目标与背景对比度低、边缘模糊、细节分辨能力差等特点，以及通常情况下的实时性处理要求，提出了一种新的有效分割方法。基于Renyi熵原理，构造了一种广义模糊熵——模糊Renyi熵；为了较快地获得分割阈值，基于混沌理论设计了一种混沌模拟退火算法，用于最佳分割阈值的搜索；把提出的模糊熵与混沌模拟退火算法相结合用于红外人体图像分割，并与几种著名的图像阈值分割方法进行了比较。实验结果表明，用该方法对红外人体图像进行分割，能得到较满意的分割结果，与其他方法相比，鲁棒性较好；对具有256级灰度的图像进行分割，其CPU耗时约为0.8 s，满足了红外人体图像分割的精确、实时性要求。

关键词: 红外人体图像分割, 模糊Renyi熵, 混沌模拟退火