K-均值聚类中心分析法实现红外人体目标分割

Human Segmentation Algorithm in Infrared Images Based on K-means Clustering Centers Analysis

针对由于不同红外成像设备参数差异以及目标周围环境影响而造成的红外目标分割阈值自动选取算法的鲁棒性差这一问题,本文从红外成像的机理出发,提出了一个新的解决方案并加以实现。首先对图像的直方图采用K-均值聚类,然后通过对聚类中心分布特点的分析,确定图像分割的阈值。该方法不需要事先对图像进行均衡和对背景分布进行假设。实验结果表明,算法具有良好的鲁棒性。

Due to the differences of various infrared cameras parameters and the surroundings influence, the robustness of the threshold auto-selection algorithm in infrared images segmentation has not been well resolved. Based on the mechanism of infrared imaging, we presented a new solving scheme and put it into practice. Firstly, the image histogram was clustered by K-means clustering method, and then the distribution characteristic of the cluster centers was analyzed in detail and the threshold for image segmentation was determined. This algorithm doesn＇t need to equalize the image before segmentation and assume the distribution of the background. Experimental result shows good robustness.