基于核函数的自适应聚类背景提取

Background Extraction Based on Adaptive Improved Clustering by Kernel Function

随着目标的检测与跟踪的广泛应用,移动物体分割问题成为研究热点,而解决这类问题的关键需为对图像背景的更新和提取。通过对目前已有的背景提取方法的研究,提出了一种新的基于核函数的背景提取方法:进行预处理,制定删选法则来自适应获取所需处理的视频序列;定义核函数来衡量模式间相似性,对2类聚类中心进行初始化;利用最小距离准则来灰度归并,采取实时的自适应阈值选择;最后通过比较满足条件的像素总数,获取所需背景。实验证明,该方法有效。

The applications of detection and tracking traffic car are paid more attention,the key part is to segment the moving target,for the purpose of solving the problem,it＇s urgent to update and extract the background,after researching the existing methods,a new algorithm based on improved clustering by kernel function is proposed,firstly,a pre-disposal formula is to get the useful video sequences,then define the kernel function and initialize the class,secondly,according to the formula,it need to merge the similar class,and take the adaptive threshold selection,the method combine the traditional clustering and on-line clustering,at last,compare the pixel＇s number which meet the condition,then choose the max number＇s class as the background,experiments show that this method is effective.