灰度序列图像中基于纹理特征的移动阴影检测  

Shadow detection based on texture features in gray sequence images

作　　者：韩延祥[1] 张志胜[1] 郝飞[1] 陈平[1] 

机构地区：[1]东南大学机械工程学院,江苏南京211189

出　　处：《光学精密工程》2013年第11期2931-2942,共12页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.50805023);江苏省"六大人才高峰"资助项目(No.2008144);东南大学优秀青年教师教学科研资助项目(No.200902);江苏省"青蓝工程"资助项目(No.2008-30-1)

摘　　要：实施移动目标检测时,由于移动阴影与运动目标具有相似的运动特性,常严重影响检测结果。本文假设视频序列中当前图像的阴影区域与对应的背景部分有相同的纹理特征,提出一种新的基于纹理特征的移动阴影检测方法。为了实现纹理特征的表达,在阴影检测过程中提出了几种纹理相似度测量方法,并通过计算当前图像与背景图像的纹理相似度值来区分当前图像中阴影区域和前景目标。另外,基于Ostu算法(大津算法)提出了一种自适应的阴影分割方法,并分别对室内和室外环境下采集的视频进行了阴影检测。定性和定量分析表明,在具有显著纹理结构的视频序列下,所提出的方法对于阴影检测有很好的效果,而对于图像模糊并伴有噪声的图像检测精度会有所降低。Moving object detection is usually effected by moving shadows from the object and it is easy to produce serious problems because the object and the shadow have the same moving characteristic. This paper proposes a method for moving shadow detection based on texture feature under the as- sumption that shadow regions in the current image have the same texture feature with a corresponding background image. For the discrimination between shadows and foregrounds, several texture similari- ty measurements are used and the shadows and foregrounds in current image are distinguished by com- puting the similarity of current image and background image. Furthermore, an adaptive method for segmentation of shadows and foregrounds is presented based on Ostu algorithm. In the end, experi- ments are carried on the video extracted from indoor or outdoor environments. Analysis indicates that the method has a excellent detection result in a video sequence with outstanding texture features. However, detection accuracy will be reduced when the image suffers blurring or noising. The obtained results verify the efficiency of the proposed method by using both of subjective evaluation and objec- tive evaluation.

关 键 词：灰度序列图像 阴影检测 OSTU算法 纹理相似度测量 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]