A Novel Binary-valued Hopfield Network Based Iterative Approach to Stereo Matching
Traditional neural network algorithms for stereo matching often suffer the poor performance and slow or premature convergence. In order to avoid these problems, a novel Hopfield network based iterative stereo matching approach is proposed. Firstly, the optimal search problem of disparity map is transformed to an iterative convergence process of binary-valued neural network, whose maximal number of neurons is the size of image. Secondly, the disparity prelabeling based on local matching method is used to initialize the weights of neural network. Moreover,according to the implicit assumption in local matching algorithms, two evaluation criteria are proposed to determine the disparity range of each pixel, which can reduce the number of active neurons in each iteration.Experiments indicate this approach is much better than traditional algorithms in performance and convergence speed.
Stereo Matching Hopfield Network Iterative Approach Energy Minimization Disparity Pre-labeling
Lili Lin Wenhui Zhou
College of Information and Electronic Engineering, ZheJiang Gongshang University,Hangzhou, Zhejiang, Department of Information Science and Electronic Engineering, ZheJiang University, HangZhou, Zhejian
国际会议
杭州
英文
795-798
2006-10-12（万方平台首次上网日期，不代表论文的发表时间）