基于自适应权值的图割立体匹配算法

Stereo Matching Algorithm Based on Adaptive Weight and Graph Cuts

摘要:

针对传统自适应权重存在计算复杂度高和传统图割匹配精度低的问题，提出一种改进的基于自适应权值和图割的立体匹配算法。首先根据灰度相似性和空间相近性重新定义能量函数的数据项，利用图像的梯度信息作为能量函数的平滑项；然后运用图割理论和 α 扩展算法进行模型求解；最后对视差图运用左右一致性检验和加权中值滤波进行视差精化。算法采用了Middlebury 网站提供的四幅国际标准立体图像对进行测试，实验表明所提算法能够得到较准确的视差图。

Abstract:

Aiming at the problem that the traditional adaptive weight has high computational complexity and low precision of traditional graph matching, an improved stereo matching algorithm is proposed based on adaptive weights and graph cuts. Firstly, the data items of the energy function are redefined according to the gray similarity and the spatial similarity. The gradient information of the image is used as the smoothing term of the energy function. Then the model is solved by using the graph theory and the α -expansion algorithm. Finally, the disparity map is used to perform the parallax refinement by using the left - right consistency check and the weighted median filter. The algorithm uses the four international standard stereo images provided by the Middlebury website to test, and the experiment shows that the proposed algorithm can get a more accurate disparity map.