关键词: 图像特征, 双层SVM, 自适应步长, 窗口融合, 纹理增强

Abstract: The shadow usually have some interference on the analysis of machine vision, the algorithms of edge extraction, object recognition and image matching, while many proposed algorithms often can’t process the texture and edge of the shadow areas well.This paper presents an approach to shadow detection and removal for the texture-consistent shadow.Shadow detection：Double-layer SVM trainer is used in this paper to detecte shadow regions by sampling colors, texture, directions and distances features of the shadow regions and lit regions between shadow edges, which can detect both shadow and lit regions corresponding to the shadow region.Shadow recovery：Adaptive steps of windows matched algorithm is used to get the optional windows and recover shadows preliminary through windows merging, then enhance texture iteratively.Finally, the Fast Marching Method（FMM） is adopted to remove slight shadow edges.The result of Opencv experiment shows that this algorithm can not only improve the accuracy of the shadow detection but also have a better removal effect on texture and edge of shadow areas.

Key words: image features, double-layer SVM, adaptive steps, windows merge, texture enhance