基于视差图优化的立体匹配算法研究

Research on stereo matching algorithm based on disparity map optimization

传统的立体匹配算法通过计算两幅图像的像素点相似程度,从而找到左图像素点和右图像素点的对应关系,常采用的是一种局部最优的方法,所以其匹配精度很低。为了提高精度,当前策略主要是将问题转化为求解能量方程,进而来对全局空间进行优化。从一个新的角度即视差图优化来研究能量方程的求解,通过对扫描线算法、动态规划算法和图割算法深刻讨论来分析算法原理与运行图,从而理解各算法优缺点形成本质与问题转化的依据,可以进一步的改进其不足,为研究新的算法打下基础。

The traditional stereo matching method is calculating the similarity of the pixels in two images and then finds the correspondence between the left picture pixels and the right picture pixels.However it is often a local optimum method and the accuracy is very low.To improve accuracy,the current strategy is to create its energy equation,and then to optimize the global space.However,it is difficult to understand the reasons for the transformation and the essence of the problem.Thus from the perspective of disparity map optimization,the optimization of the global energy equation is discussed by using the scanline optimization algorithm,dynamic programming algorithm and graph cut algorithm.Through a deep analysis of their nature and the diagram when the algorithms are running,the advantages and disadvantages of each algorithm are analyzed,and their shortcomings are further improved,thus laying the foundation for new algorithms.