摘要：

远红外图像中人体目标分割阈值自动选取算法的鲁棒性较差。为此，从远红外图像的成像机理出发，提出一种改进的K均值聚类中心分析法。当所属类别不同时，聚类前呈线性分布的聚类中心会在聚类后明显转折。根据该特点，将聚类后待测类别的实际聚类中心值与理论聚类中心预测值的绝对差值作为测度函数，选择转折点并确定图像分割的阈值。实验结果表明，该算法具有良好的鲁棒性与抗噪性。

关键词: 红外图像分割, K均值聚类中心分析, 转折点选取, 行人探测

Abstract:

Aiming at poor robustness of the threshold auto-selection algorithm in far-infrared images segmentation, an improved K-means clustering centers analysis algorithm based on the mechanism of far-infrared imaging is researched in this paper. According to the character that the cluster centers had a linear distribution before clustering and had a clear turning point after clustering when they belongs to different categories, the absolute difference between the practical cluster centers value and theoretical cluster centers predicting value of a category under test is taken as the measurement function to select the turning point, thus the threshold for image segmentation was determined. Experimental result shows good robustness and anti-noise performance of the algorithm.

Key words: infrared image segmentation, K-means clustering centers analysis, turning point selection, pedestrian detection

中图分类号: 

• TP391.4