基于线性滤波的树结构动态规划立体匹配算法

A Tree Structure Dynamic Programming Stereo Matching Algorithm Based on Linear Filtering

传统的动态规划立体匹配算法能有效保证匹配精度的同时提高运行速度,但得到的视差深度图会出现明显的条纹现象,同时在图像弱纹理区域以及边缘存在较高的误匹配.针对该问题,提出了一种新的基于线性滤波的树形结构动态规划立体匹配算法.算法首先运用改进的结合颜色和梯度信息参数可调的自适应测度函数构建左右图像的匹配代价,然后以左图像为引导图对构建的匹配代价进行滤波;再运用行列双向树形结构的动态规划算法进行视差全局优化,最后进行视差求精得到最终的视差图.理论分析和实验结果都表明,本文的算法能有效地改善动态规划算法的条纹现象以及弱纹理区域和边缘存在的误匹配.

The traditional dynamic programming stereo matching algorithm can effectively guarantee the precision of matching and improve the running speed; but the depth of the parallax figure has the obvious stripes phenomenon, at the same time the low texture region and edge of the image have higher mismatch. For these problems, the paper proposes a new tree structure based on the linear filtering dynamic programming stereo matching algorithm. The algorithm firstly uses an improved adjustable parameters adaptive measure function to combine the color and gradient information of the matching images. Secondly, it uses the left image to guide the figure to filter the price of stereo matching. Thirdly, it utilizes the two direction simple tree structure dynamic programming optimization; and finally uses the parallax refinement method to get the final parallax figure. Theoretical analysis and experimental results have showed that the proposed algorithm can not only effectively eliminate the stripes phenomenon of dynamic programming algorithm but also improve the mismatch of the low texture area and the edge of the image.