结合图像梯度和亮度的并行信任传播算法

Parallel Belief Propagation Algorithm Conjugating Gradient and Intensity

对立体匹配问题建立马尔可夫随机场模型,使用并行的多尺度信任传播算法求解马尔可夫随机场的能量最小化问题.在传统串行算法基础上利用CUDA技术实现了并行计算,并结合图像的梯度和亮度信息计算能量函数的数据项,平滑项采用两个相邻像素视差的绝对差度量.以标准的Middlebury立体数据集做为输入,实验结果表明:算法具有很好的实时性能,运行时间远小于传统的串行算法,深度图结果较优.

Stereo matching is critical technology in vision measurement.MRF models are established to do with stereo problem.A parallel multi-scale belief propagation algorithm is used for MRF energy minimization and generating disparity map.Parallel algorithm is implemented based on traditional sequential algorithm with CUDA technology.In energy function,data term is conjugated with Gradient and intensity of images,smooth term is measured with the absolute difference of disparities between two adjacent pixels.With standard Middlebury stereo data sets as input,experiments show that the proposed algorithm has good real-time performance,running time is much less than the traditional sequential algorithm and the generated disparity map is excellent.