基于Kmeans和图像熵聚类的热红外目标检测算法

Infrared target detection algorithm based on image entropy and cluster Kmeans

针对以海洋为背景的热红外图像目标检测存在的海洋海杂波的非平稳特性、非线性特性问题,以及目标背景相关性大而对比度小等问题,对两幅实拍红外船舰图像进行了实验,提出了一种快速有效的热红外目标检测算法。该算法采用表示图像灰度空域分布状态不确定性量度的图像熵方法,利用滑窗方法遍历整幅图像,求得了局部熵图像,从而确定了目标的粗略位置;通过用最大类间差法将局部熵处理后图像进行了自适应的二值分割,将目标和背景最优化地分离,并且结合改进的Kmeans聚类算法,通过循环所有目标点找出了其在聚类图像中的聚类标识,结合所有该聚类的像素点,提取出了完整的目标及其轮廓。研究结果表明,该热红外目标检测算法速度快,性能良好,在将目标完整地提取出来的同时可以很好地保留目标的轮廓。

Aiming at the infrared target detection in the sea background with non-stationary and non-linttar cluster, a novel infrared target detection algorithm was proposed after the experiment in two real infrared ship images. The local image entropy was generated by sliding windows through out the whole image to locate the targets roughly. The Otsu method was adopted to realize self-adaptive binary zation segmentation and the whole target and outline were extracted combining with Kmeans cluster algorithm. The results indicate that the targets are extracted effectively as well as the outline with speed and performance.