一种有效的图像阴影自动去除算法

An effective shadow removal approach

视频内容分析要求比较精确的目标检测方法。常用的背景减方法在检测运动前景时也会检测到运动目标投射的阴影,将阴影区域误认为目标区可能造成运动目标粘连或者目标区域几何变形,影响后续内容分析结果,因此去除阴影对于提高后续内容分析的准确性提供了保障。本文提出一种基于颜色统计特性的阴影去除方法。首先利用背景减的方法得到包含阴影的候选目标区域。进一步,分析候选目标区域和背景在YCbCr颜色空间的差值统计特性,发现阴影区域有很强的规律性:色度分量与背景区域一致性很高,亮度分量有固定的差。根据上述规律,设计算法去除阴影区域。与现有方法的对比实验结果表明,本文方法能够很好的去除阴影区域,同时又保持了前景目标区域的完整性。

Effective object detection is important in video analysis.The typical approach such as background subtraction can not differentiate between the object and shadow.It results in the connection of multi objects and deformation of segmented object,which will influence the accuracy of the follow-up content analysis results.Thererfore,it is necessary to remove the shadow in object detection.In this paper we propose a color statistics based shadow removal approach.First,the background subtraction is used to get the candidate regions,which generally include the real object and the shadow.Next,we compute the difference between the foreground image and background image for the pixel in the candidate regions in the YCbCr color model.Then we analyze the statistics of foreground regions and shadow regions.And we find that the differences of Cb and Cr are much smaller in shadow regions than the foreground regions.And the luminance difference is in a determined range.By the analysis,the approach is designed to remove the shadows.The compared experimental results show that our proposed approach can remove the shadow region and keep the integrity of the foreground object.