基于视差范围自动提取的视差图优化算法研究

RESEARCH ON DISPARITY MAP OPTIMISATION BASED ON AUTOMATIC DISPARITY RANGE EXTRACTION

基于图割理论的立体匹配是从同一场景不同角度的两幅或多幅图像中得到视差图,进而求得场景的立体深度,为三维重建、视觉测量等提供有用的信息。然而,视图视差范围的选择往往被忽视,提出一种基于图像分割及GPU SIFT算法的视差范围快速自动提取算法,得到的视差范围值应用于立体匹配算法中视差范围的取值;并针对图割方法进行立体匹配时存在的正面平行表面缺陷的问题,提出一种最小二乘法的迭代改进方法,对立体匹配得到的视差图进行优化处理。对Middlebury中的标准数据进行测试的结果表明,该算法能准确地提取视图的视差范围,并有效提高各类平面与曲面的视差精度。

Stereo matching algorithm based on graph cuts theory produces disparity images from two or more images of the same scene in different visual angles,from which the stereo depth of scene can be further obtained,this brings available information to 3-D reconstruction and vision measurement.However,the choice of the disparity range is often overlooked.In the paper we present a fast disparity range auto-matic extraction algorithm which is based on image segmentation and GPU SIFT algorithm,the value of disparity range derived is applied to the value assigning of disparity range in stereo matching algorithm.Moreover,we present an iterative improvement method of the least squares for solving the problem of front parallel surface defect happened in stereo matching with graph cuts,and carry out optimised processing on dis-parity image obtained in stereo matching.It is demonstrated by the result of the test with standard data in Middlebury that the proposed algo-rithm can extract the disparity range of visual image accurately,and improves the accuracy of the disparity of various flat and curved planes ef-fectively.