采用纹理分析的运动人体阴影分割方法

研究采用纹理分析的方法来消除由光照产生的人体阴影,进行人体阴影分割,首先采用混合高斯模型对运动人体前景及背景进行提取,主要研究采用多尺度小波分解来提取运动人体背景及前景纹理及颜色特征,建立基于像素的纹理及颜色特征向量,并采用支持向量机对特征向量进行分类,如前景像素点特征向量与对应的背景像素点特征向量一致时,可判定为阴影区域。结果表明:采用此纹理分析方法可很好地分割出阴影区域,并且光照变化及采集图像时的颜色偏差对阴影提取几乎没有影响。

A Method for Moving Human Shadow Segmentation Using Texture Analysis

In this paper,texture analysis method is presented to segment and remove the human’s shadow produced by the light.The method is composed of three steps:firstly,using the Gaussian mixture model to extract the foreground and background from the Human motion sequences,secondly,using wavelet decomposition to extract texture features and color features of the foreground and background and establish the eigenvector based on the pixels,finally,using SVM to classify the eigenvector and if the foreground’s eigenvector and background’s eigenvector are the same,the pixel is the shadow.The results showed that using the texture analysis method shadows are segmented clearly and it can do well in various environments,for example the illumination changes or color deviation.