基于像素块最大后验概率的视频对象分割方法

刘少华; 张茂军; 智利丁

doi:10.11834/jig.20090122

图像分析和识别 | 浏览量 : 0 下载量: 352 CSCD: 0

PDF
导出
分享
收藏
专辑

基于像素块最大后验概率的视频对象分割方法
Video Object Segmentation Based on MAP of Pixel Blocks
2009年14卷第1期页码：125
纸质出版：2009
DOI： 10.11834/jig.20090122
稿件说明：

移动端阅览

刘少华, 张茂军, 智利丁. 基于像素块最大后验概率的视频对象分割方法[J]. 中国图象图形学报, 2009,14(1):125. DOI： 10.11834/jig.20090122.

LIU Shaohua, ZHANG Maojun, ZHI Liding. Video Object Segmentation Based on MAP of Pixel Blocks[J]. Journal of Image and Graphics, 2009, 14(1): 125. DOI： 10.11834/jig.20090122.

摘要

提出一种基于像素块最大后验概率的视频运动对象快速分割算法。首先根据连续3帧视频序列构建参考帧，计算基于像素块的差分图像；然后用Gibbs随机场模型描述基于像素块的差分图像，通过最大后验概率方法分割运动对象块；最后使用基于颜色梯度的snake模型提取运动对象轮廓。实验结果表明，该算法能避免视频对象的缺失，降低图像噪声的影响，减少误分割，同时具有较快的计算速度。

Abstract

A video object segmentation method based on MAP (maximum a posteriori) of pixel blocks is proposed. Firstly

a reference frame is built from three continuous frames and a pixel block-based difference image is calculated. Secondly

a pixel block-based difference image is described by Gibbs random field model

the moving object blocks are detected by MAP. Last

the contour of the moving object is extracted by the snake method. The experiments results show that the algorithm can avoid the losing and mistakes with a good speed.