传统红外热像存在对比度低、边缘模糊等不足而使目标区域分割困难, 红外偏振热像能够凸显边缘和轮廓特征, 因此在环境监测、**侦察、工业无损检测等领域得到广泛的应用, 但如何进行红外偏振热像分割目前研究较少。为此, 本文提出了一种基于 Tsallis熵的红外偏振热像分割算法。首先通过 Tsallis阈值对偏振方位角热像进行初分割, 然后以最小化初分割热像交集与并集误差率优化 Tsallis指数, 再利用指数优化后的 Tsallis阈值对偏振方位角热像进行优化分割并通过连通域检测去除误分割得到二次分割图, 最后以二次分割图交集区域为种子区域、并集区域为边界, 通过区域生长法得到最终分割热像。实验结果显示, 本文算法相对最小 Tsallis交叉熵法、Otsu法和模糊聚类法错分区域小, 在主观视觉效果和区域间对比度、形状测度评价指标上有较大的改善,能够更准确地分割出目标。

There are some deficiencies in traditional infrared thermal images, such as low contrast and blurred edges, making segmentation of the target region difficult. Infrared polarization thermal imaging can clearly highlight the edges and contours of the observed objects. It has been effectively applied in environmental monitoring, military reconnaissance, industrial nondestructive testing, and other fields. However, there are few studies on infrared polarization thermal image segmentation. In this work, a novel infrared polarization thermal image segmentation algorithm based on Tsallis entropy is proposed. First, the polarization azimuth thermal image is segmented with a Tsallis threshold. Second, the Tsallis index is optimized by minimizing the intersection and union error rate of the initial segmentation thermal image. Subsequently, the polarization azimuth thermal image is segmented by utilizing the exponentially optimized Tsallis threshold, and the quadratic segmentation image is obtained by moving false segmentation through connected domain detection. Finally, taking the intersection area as the seed region and taking the union region as the boundary, the final segmentation thermal image is obtained through the regional growth method. The experimental results show that the algorithm proposed can diminish the false segmentation region. It can improve the evaluation index of the subjective visual effect, interregional contrast, and shape measurement and segment the target accurately compared with the minimum Tsallis cross-entropy, Otsu, and fuzzy clustering methods.