The Study On Stereo Matching Based On Color Segment And Plane Consistency

Three-dimensional video system is rapidly developed with the growingspiritual needs of human social activities. Traditional visual media can onlyprovide two-dimensional visual information, but the three-dimensional videosystem has a stronger, more comprehensive explanatory power to describe thescene. There are two ways to get depth information. Firstly, the pixel depthinformation can be obtained by the corresponding point, which isthree-dimensional image parallax match. Secondly, we can use the depthsensor to measure the depth of the natural scenery directly.Disparity estimating is an ill-posed problem. In recent years, the imagesegmentation methods are often used to optimize disparity because disparityhas a strong correlation in the same segmented region. Based on disparitycontinuity and plane consistency assumptions, we proposed SBRDF (SegmentBased Reliable Disparity Fitting) to estimate disparity: using two localdisparity estimation algorithms to obtain reliable disparity by left to rightcheck and cross-consistent check. Then define the center disparity for eachsegmented region to eliminate noises. Least squares method is used to get thefinal disparity result at last. Compared with the traditional segmentation basedmethods, SBRDF is not involved with interactive optimization steps.Experimental results show SBRDF guarantee the accuracy of disparitymatching and effectively improve the efficiency.The graph cut algorithm is a common global optimization algorithm, andits energy function based on different mathematical models is very important.The energy function of traditional graph cut algorithm consists of two terms,the first term is the data function based on local algorithm to establish dataconsistency, and the second term is the smoothness function betweenneighbor pixels to enforce smoothness. However, the data term based on localalgorithm may have error because of occlusion and violation of smoothness assumption. SBRDF can effectively inhibit the proliferation of errors betweensegmented regions, the edge of the disparity is relatively clear. Therefore, wepropose a plane consistency constraint graph cut (PCGC) method to estimatedisparity referring to SBRDF. The PCGC has strong fault tolerant ability toimprove the result of SBRDF while the result of PCGC is much better thanGC.Finally, we study the KINECT depth image. Segment plane consistencyinpainting (SPCI) algorithm is proposed to inpaint holes on KINECT depthimage. The KINECT can provide color images of the scene and thecorresponding depth images. With the camera parameters, disparity image anddepth image can be transformed into each other. So the method of disparitymatching algorithm based on segmentation and can be apply to inpaintingdepth images, since there is a strong correlation between segment regions.Because the SPCI is based on regions, the boundary of image after inpaintingis much better.