基于自适应窗和行列双动态规划的改进立体匹配算法

摘要

An improved stereo matching algorithm based on adaptive⁃window and determinantal dual dynamic programming is proposed in this paper. RGB space is imported into the adaptive window stereo matching algorithm as an initial disparity map to improve matching accuracy,and then a global energy optimization algorithm based on determinantal dual dynamic program⁃ming is proposed to improve apparent defective stripes caused by dynamic programming. The energy minimization model is used to solve the optimization problem,during which the corresponding data items is introduced for different reward values. The different reward mechanism is given based on the change of the initial disparity in the direction of column to make the disparity map clo⁃ser to the actual disparity. Experimental results show this algorithm can effectively reduce the matching error rates in depth dis⁃continuities,low⁃textured regions and apparent defective stripes,and improve the image matching quality.%　　提出了一种基于RGB空间的改进的自适应窗口和行列双动态规划的快速立体匹配算法。该算法首先将RGB空间引入自适应窗立体匹配算法中作为初始匹配结果，有效提高了匹配的精度；同时针对传统动态规划算法复杂度较低、实时性高，但容易产生条纹瑕疵的特点，提出了基于行列双动态规划的自适应全局能量优化算法。采用能量最小化模型对最优化问题进行求解，在基于行列动态规划的求解过程中引入对能量函数的数据项进行奖励的策略，根据列方向上初始视差的变化情况给予不同的奖励机制，从而使得到的视差图更接近于真实视差图。实验结果表明，该算法可以有效降低视差图在边缘区域和低纹理区域的误匹配率，减少条纹瑕疵，提高图像匹配的质量。