Research On The Approach Of Shadow Detection For Single Outdoor Images

The approach of shadow detection had been an important topic in many computervision research areas, such as image segmentation, target recognition, tracking and contentunderstanding. How to accurately detection shadow in images is one of the key links tosolving the above problems of computer vision. Based on analysis of the present situationof the domestic and foreign research, this paper do research on the approach of shadowdetection for single outdoor images by using support vector as technical foundation andcomprehensivly using many aspects of theories knowledge, such as image segmentationand content understanding.First, several image preprocessing methods commonly used in the process of shadowdetection are introduced, and the technology of support vector machine which is used forclassification is also introduced.Second, an approach of shadow detection which is based on boundary is introduced,and the approach of shadow detection based on content understanding for outdoor imagesis proposed. In this approach, the shadow are classified to ground shadow and verticalshadow according to the results of content understanding, and vertical shadow are verifiedagain according to its position to gain all shadow in image.Third, an approach of shadow detection based on region is introduced, and shadowdetection for outdoor images based on region segmentation and combining with gradientdirection feature is proposed. In this approach, a part of shadow regions and the regionswhich have the same material and illumine with shadow regions are firstly detected byclassifiers. According to the idea that all of the neighbor regions which have the samematerial and illumine with shadow must be shadow, then all shadow regions can bedetected.Finally, experimental results show that the proposed methodes are feasible andeffective, and the analysis of experimental results are made.