借鉴生物先天性免疫与适应性免疫的协调作用机制, 综合考虑中波红外图像的光谱成像机理和频域模板统计特征, 提出一种免疫模板聚类目标提取算法。 借鉴先天性免疫对抗原表面分子模式的识别作用, 以最大类间方差, 将模糊中波红外图像初分割为目标像素集、 背景像素集和模糊像素集; 借鉴先天性免疫的特征提呈作用, 提取中波红外图像模糊像素的频域模板特征, 将图像的像素灰度特征空间映射为频域模板特征空间; 基于提呈得到的模板特征, 对模糊像素集进行适应性免疫聚类, 将模糊像素划分为目标像素或背景像素。 用手部痕迹的模糊中波红外图像进行实验, 并与经典边缘检测模板法和传统区域模板法进行了效果比较和定量评价, 结果表明免疫模板聚类算法的目标提取率、 与参考标准的重合度、 绝对误差率均优于现有模板方法, 能够有效实现模糊中波红外图像的目标提取。

Extracting targets from a blurred midwave infrared image is a challenging task due to the fuzziness of the image. Inspired by the coordination mechanism between biological innate immunity and adaptive immunity, an immune template clustering targets extraction method is proposed, which based on imaging mechanism and template statistical property of midwave image. Firstly, by learning from the recognition function of innate immunity and maximizing the between-cluster variance, a midwave blurred infrared image is segmented into a target pixel set, a background pixel set and a blurred pixel set. Secondly, according to the presentation function of innate immunity, the frequency domain template features of pixels in midwave blurred infrared image are extracted. Finally, adaptive immune clustering is completed for the blurred pixel set based on frequency domain template feature, in order to divide each blurred pixel into target pixel or background pixel. Experimental results show that the proposed algorithm can extract targets from a midwave blurred infrared image efficiently. Compared with classical edge template and conventional region template methods, the immune template clustering method has better extraction efficiency, absolute error rate and coincidence degree with ground truth.