基于马尔可夫随机场的单目图像深度估计  

Depth Estimation from Monocular Images Based on MRF Model

作　　者：张蓓蕾[1] 刘洪玮[1] 

机构地区：[1]东华大学信息科学与技术学院

出　　处：《微型电脑应用》2010年第11期49-50,59,共3页Microcomputer Applications

基　　金：国家自然科学基金(60502042);上海市启明星基金(06QA14003)资助

摘　　要：图像深度获取是机器视觉领域活跃的研究课题。将图像深度估计问题归结为模式识别问题,以单目图像深度为待分连续模式类,在多尺度下对图像块提取绝对和相对深度特征,选择表征上下文关系的MRF(Markov Random Field)-MAP(Maximum a posteriori)方法,建立拉普拉斯模型,表述某图像块的深度和其邻域深度之间的关系。实验得到了某一类单目图像对应的深度图像,证明了该算法的有效性。Depth Estimation is an active research area in computer vision.This paper viewed this problem as an issue of Pattern Recognition.The depth of monocular images is the pattern to be classified.This paper combined absolute and relative features of monocular Images in multi-scale,and modeled the relationship between depths and features and the relation between depths at different patches in the image based on Laplacian MRF-MAP Mode.And our algorithm is able to receive depth-maps for a certain kind of monocular images.

关 键 词：单目图像 深度图像 马儿可夫随机场 最大后验概率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]