基于Kinect的自动视频抠像算法  

Automatic video matting algorithm based on Kinect

作　　者：夏倩[1] 许勇[2] 夏玉勤[3] 

机构地区：[1]贵州民族大学传媒学院,贵州贵阳520025 [2]华南理工大学计算机学院,广东广州510510 [3]贵州大学计算机科学与工程学院,贵州贵阳520025

出　　处：《计算机工程与设计》2015年第5期1299-1303,共5页Computer Engineering and Design

基　　金：贵州省科学技术基金项目(LKM[2012]11)

摘　　要：针对传统的视频抠像算法交互频繁、计算复杂度高等问题,利用从Kinect获得的深度图信息、彩色信息以及视频中运动信息,提出基于Kinect的自动视频抠像算法。采用改进的三帧间差分法检测视频中的感兴趣区域(ROI),根据深度图信息改进自动种子增长区域算法,估算粗略的前景掩膜;通过数学形态学和逻辑操作去除粗前景区域的杂点,自动生成大致的三分图;采用改进的Shared Matting算法,得到精确的视频抠像。实验结果表明,该算法避免了人工交互,精度高且速度快。To solve the problems of frequent human interaction,complex computation and bad robustness of the traditional video matting algorithms,an automatic image matting algorithm based on Kinect was presented.Color,motion and depth information retrieved from Kinect were utilized.Firstly,regions of interest （ROI）of the video were detected using three frames difference of image.Then the region-growth process with the depth information was used to estimate a foreground mask.Secondly,noises of the crude foreground mask were removed through the mathematical morphological operations and logical operations to generate a trimap.Finally,good matting effects were obtained using the improved Shared Matting algorithm.Experimental results show that the algorithm has not only no human interaction,but also high precision and fast implementation.

关 键 词：感兴趣区域 前景掩膜 三分图 视频抠像 深度图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]