视频序列中的一种微动目标提取与背景替换算法  

AN ALGORITHM OF MICRO MOVING TARGET EXTRACTION AND BACKGROUND REPLACEMENT IN VIDEO SEQUENCES

作　　者：安世全[1] 文倩云 孟琦[1] 

机构地区：[1]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《计算机应用与软件》2016年第11期108-111,共4页Computer Applications and Software

基　　金：重庆市教委科学技术研究项目(KJ1402001);重庆市科委基础与前沿项目(cstc2014jcyjA1347);重庆市高校优秀成果转化项目(KJZH14219)

摘　　要：为了实现视频图像中微动目标的背景替换,需要解决微动目标准确提取这一难题。首先利用Adaboost分类器训练肤色样本库,获得肤色与非肤色数据,并确定肤色的阈值范围,得到肤色检测二值图;然后将肤色检测二值图和Otsu分割的微动目标二值图融合得到微动目标序列图像的粗二值图;经过形态学、孔洞填充等处理过程得到完整的二值图模板,最后完成视频图像的背景替换。仿真实验表明,所提出的算法可以较好地分割出微动目标并实现背景替换。In order to realise the background replacement of micro moving targets in video images,there has the need to solve the problem of accurate micro moving targets extraction. In the paper we first use Adaboost classifier to train skin colour sample database to obtain the data of skin colour and non-skin colour,and determine the threshold range of skin colour,as well as get the binary image of skin colour detection.Then we fuse the binary images of skin colour detection and the binary images of micro moving targets segmented by Otsu to get the rough binary image of image sequence of micro moving targets. After that,we obtain the complete binary image templates through the treatment processes of morphology and holes filling,and at last accomplish the background replacement of video image. Simulation experiments show that the proposed algorithm can well segment the micro moving targets and realise background replacement.

关 键 词：OTSU ADABOOST分类器 肤色阈值 形态学 背景替换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]