灰度序列图像中结合互相关法与梯度特征的阴影去除算法

Shadow Removal Algorithm by Combining Cross-correlation and Gradient Feature in Gray Sequence Images

在视频监控场景下的目标检测中,运动的阴影会对所检测目标的准确性造成不利影响.为了去除运动阴影的干扰,提出了一种结合灰度比值的四方向梯度与归一化互相关(NCC)特征的阴影去除算法.首先在灰度序列图像中计算前景与其对应背景灰度的比值;其次根据阴影区域的相邻像素灰度比值变化改变很小,通过计算灰度比值的四方向梯度均值来判断阴影并加以去除;最后为了避免运动目标被误去除,考虑到运动目标与背景的相关性差异远远大于阴影与背景之间的相关性差异,结合归一化互相关特征来保留目标,以确保运动目标的完整性.定性和定量的实验结果分析表明,该算法在阴影去除率和保持目标完整性方面优于其他阴影去除算法.

Moving shadows have adverse effects on the accuracy of the detected target while detecting moving target in video surveillance scene. In order to eliminate the interference of the moving shadows,a shadow removal algorithm is proposed by combining the four-direction gradient of the grayscale ratio and the feature of the normalized cross correlation(NCC)in this paper. Firstly,the grayscale ratio of the foreground pixel to its corresponding background pixel is calculated in gray sequence images. Secondly,because the grayscale ratio among a pixel ’ s neighborhood in shadow region has a small change,the calculated four-direction gradient mean can be applied to detect and remove shadow. Finally,in order to avoid that the moving target is removed as shadow and ensure the integrity of the detected target,the feature of the NCC is employed to keep the integrity of moving target because difference of correlation between the moving target and its corresponding background is much larger than that between shadow and its corresponding background. The qualitative and quantitative analysis of experimental results show that the algorithm presented in this paper is superior to other similar shadow removal algorithms in shadow removal ratio and keeping the integrity of the detected target.