视觉显著性的自适应权重立体匹配算法  

An adaptive weight stereo matching algorithm based on visual saliency

作　　者：王冲[1] 潘晨[1] 高浩然[1] 

机构地区：[1]中国计量大学信息工程学院,浙江杭州310018

出　　处：《中国计量大学学报》2016年第4期435-440,479,共7页Journal of China University of Metrology

基　　金：浙江省自然科学基金资助项目(No.LY13F010004)

摘　　要：针对传统算法大多忽略人眼特性,获得的视差图与人眼真实感受之间存在一定差异的问题,提出了一种符合人眼视觉特性的自适应权重匹配算法.该算法首先引入视觉显著性特征,然后对像素权值分配进行改进并提出新的匹配代价度量准则,最后采取左右视差图融合的方法获得最终视差图.相关图像实验表明,改进算法很好地解决了遮挡问题,可精确描述边缘和细节视差;相对于原算法有较大程度的性能提升.The traditional stereo matching algorithm rarely considers the characteristics of human eyes,so there is difference between the disparity map and the subjective human feelings.We presented a stereo matching algorithm based on adaptive weight.Firstly,the distinguishing visual images were introduced into the stereo matching algorithm to improve the original algorithm.Then,we proposed a new matching cost criteria.Finally,according to the left and right disparity map fusion,we obtained a disparity map in accordance with human perceptions.Related image experiments showed that the improved algorithm solved the problem of occlusion and accurately described the edges and details of the disparity map.It has better performance compared with the traditional methods.

关 键 词：立体匹配 自适应权重 显著性检测 相似性度量 视差图融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]