中文摘要：

为了改善立体匹配算法在低纹理和深度跳变区域的匹配性能,提出了一种改进的置信度传播立体匹配算法.首先利用均值漂移算法对图像进行彩色分割,然后通过自适应权重算法计算匹配代价并获取初始视差图,再利用匹配代价可信度检测和左右一致性校验将初始匹配结果按照可靠度分类,最后在全局优化的过程中分别通过可靠度分类和图像分割结果来指导置信度传播方向和范围的选择,从而优化传播路径,提高匹配性能.将该算法应用于标准库图像中可以提高低纹理和深度跳变区域的匹配精度,得到边界清晰、稠密光滑的视差图,同时,将其应用于实拍图像的三维重建系统中能够得到生动逼真的立体模型,证明了该算法的有效性和实用性.

英文摘要：

In order to improve the matching performance in low-texture and depth-discontinuous regions,an improved belief propagation-based stereo matching algorithm is proposed.First,the mean-shift algorithm is applied in the color image segmentation.Secondly,the adaptive support-weight approach is used to calculate the matching cost and initialize the disparity map.Then,the pixels are classified by different reliabilities according to matching cost confidence measure and mutual consistency check.The results of image segmentation and pixel classification are used to select the direction and range of the improved belief propagation in the global optimization.Using the new technique,a better path of propagation is established and the matching performance is improved.Experimental results show that the method is efficient and can produce dense disparity maps with high accuracy even in low-texture and depth-discontinuous regions when applied in Middlebury standard images matching,and the method can also generate vivid stereo models when applied in the 3D reconstruction of the image capture in the real world.