基于置信度传播和色度分割算法的深度估计

Depth Estimation of Algorithm Based on Belief Propagation and Color Segmentation

为了提高图像序列深度估计的质量,提出了一种基于置信度传播和色度分割的全局匹配算法。首先,构造了包含匹配误差项和平滑性假设的能量函数,通过置信度传播算法来求取初始视差图序列。然后用均值漂移算法对每一帧进行色度分割,对每个色度分割区域分别进行全局匹配,得到新的视差图。最后,构造包含对极几何约束的新能量函数,使用置信度传播算法进行全局匹配和迭代优化,获取最终视差图序列。实验结果表明,文中算法可以得到高质量的深度图,能够改善图像噪声、弱纹理和物体遮挡等问题。

In order to improve the quality of the depth estimation of the image sequence,a global matching algorithm based on belief prop-agation and color segmentation is proposed. First,the algorithm constructs a matching energy function that contains the matching error term and smoothness assumptions. Estimate the initial disparity map sequences through using the belief propagation algorithm. Then,the each frame is segmented into several color areas by employing the mean shift algorithm. For each of the divided color segmentation re-gions,respectively processing the global matching to obtain the new disparity maps. Finally,create a new energy function that contains a epipolar geometric constraints,to obtain the final disparity map sequences via belief propagation algorithm using global matching and iter-ative optimization. Experimental results demonstrate that the algorithm can get high-quality depth maps and improve image noise,texture-less and occlusions problems.