Study On Image-based View Interpolating

This thesis concentrates on the research of the techniques and algorithm of view interpolation based on images, according to multiple view geometry in computer vision. First, some fundamental theory in multiple view geometry such as single view geometry, two view geometry and camera calibration, are introduced in this thesis. Then the basic principles of the techniques of image interpolation based on images are discussed. At last, the auto-matching algorithm put forward in the thesis is expatiated and it is testified in real images.So far, the algorithms of view interpolation based on images are involved in the estimation of multi parameters and image matching under multiple view geometry. The operation procedure is based on the inner interpolating of the position and color of image pixels. They include two stages. One is to establish the corresponding relation of images. The other is to get the position and color of interpolated views.In view interpolation, the most important and difficult stage is to establish the corresponding relationship. There are many methods to establish correspondences. In this paper, an auto-matching algorithm is presented. There are four steps in this algorithm. First, two original images are prewarped to make their view-spots satisfy the relation of pure translational motion. Second, the edge points on the scan-lines of the first prewarped image are detected and the corresponding points on the same scan-lines of the other prewarped image are matched. Third, with these corresponding edge points, the scan-lines of the two images are segmented. In the end, more exact corresponding relation of these points are obtained under the use of the affine transformation and the auto-corresponding of the images are finished.The second stage is to get the position and color of the interpolated view. Assuming that the illumination corresponding with origin views doesn't change, the colors of the interpolated views and that of origin views will not change too.At last, the thesis introduces the application of the auto-matching algorithm in the interpolating experiments of the images of a computer display device and a man's face. The experiments proved that the auto-matching algorithm is quite good at the speed and precision of the interpolation.