Research On Algorithm Of Stereo Correspondence Based On Graph Cut Theory

Algorithms of stereo correspondence are of great importance in binocular stereovision, and produce dense stereo disparity images. Dense disparity images is the basisof vision measurement and3D reconstruction or other applications. These algorithmshave one thing in common: stereo image pairs should be under the constraint ofEpipolar Geometry. A calibration is applied to get the intrinsic and extern matrix ofcameras. Among algorithms of stereo match proposed, the algorithm based on GraphCut works fine, yet this method consumes a lot of time.This paper mainly works on stereo correspondence based on Graph Cut, our mainworks list below:First, due to inaccuracy of images captured by cameras, we need to know whatthe matter is; in fact, these issues include focus and distortion coefficences of lens.Given these parameters we can get corrected images, this is the main aim of CameraCalibration. We experiment with Zhang’s algorithm to calibrate cameras. Acalibration method is given in this paper; a line remains a line if it is projected to animage, according to the principle of Geometry projection. Using this method, we canget the distortation information. Next we get the intrinsic and extern matrix of thecamera.Second, we propose a quick stereo correspondence method, trying to reduderunning time. Wavelets transformation is introduced to compress images. Image spaceand maximium disparity scale down, thus computing complexity scales down as well.During the process of wavelets transformation, disparity information is lost; we needto recover it using scaling and interpolation. Finally, a refinement is given to refinethe result disparity image. Experiment shows the robust and effective of our method.