Video Object Segmentation with Adaptive-K Gaussian Mixture Model and Texture

摘要：

Extracting moving objects in videos taken from a static camera is a basic technique for computer vision applications. However dynamic cast shadows are often classified as fake objects. A new video segmentation algorithm, which combines an improved Adaptive-k Gaussian Mixture Modcl(AKGMM) and local texture analysis, is proposed in this paper. The improved AKGMM is established to detect a set of moving regions. The extracted moving regions may contain potential shadow points. Shadow regions exhibit same textural characteristics in each frame and in the corresponding adaptive background model. The similarities between textures and chrominance angle are analysed for potential shadow region in order to finally eliminate shadow regions.

关键词： Shadow suppression Adaptive-K Gaussian Mixture Model Local Binary Pattern

作者: Hao Zhou Yu Gao Judong Li Xuejie Zhang

作者单位: Information school Yun Nan University Kun Ming Yun Nan Province

会议类型: 国际会议

会议名称: 2010 International Conference on Circuit and Signal Processing(2010年电路与信号处理国际会议 ICCSP 2010)

会议地点: 上海

会议语种:英文

页码: 408-411

在线出版日期: 2010-12-25（万方平台首次上网日期，不代表论文的发表时间）