Extraction Corresponding Feature And Image Synthesis Of New Viewpoint

Image-Based Rendering (IBR) technique is an alternative strategy for achieving Virtual Reality (VR), and it has become an important research area in Computer Graphics in recent years. IBR is a new technique that combines many results in Computer Graphics, Computer Vision, Image Processing and Photogrammetric. It replaces the traditional geometric modeling with the real world images, to make the rendered image real and natural and the rendering time independent of the scene complexity. The image synthesis of new viewpoint and some related key technologies in the IBR are researched in this dissertation.The extraction of the features is the basic technique of Image Processing. For the need of the matching and image synthesis the method of extraction can detect the different features of corners and edges and precision of the corner location is relatively good. The model of phase congruency is studied. The classical moment analysis is applied to phase congruency in all orientations. The resulting corner and edge operator is highly localized and has responses that are invariant to image contrast.In some image synthesis techniques, the corresponding features in the images are selected manually. Automation of the point correspondence is difficult. Finding correspondence between two widely separated views is an important open issue in computer vision. This dissertation realizes automatic extraction of the epipolar geometry in wide baseline images pairs, and introduces robust methods for automatically matching features in images corresponding to the same physical feature on an object captured from two arbitrary viewpoints. This approach can be applied to the image synthesis to realize its automatism. A sufficient amount of correspondences between the images is acquired by matching local invariant regions. 'Invariant regions' are image patches that automatically deform with changing viewpoint as to keep on covering identical physical parts of a scene. Such regions are then described by a series of invariant features, which makes it relatively easy to match them between views and under changing illumination. Robustness to false matching is obtained by applying RANSAC.A good line drawing can provide much the information that might be contained in a photograph of the same scene, and in doing so only requires a small fraction of the data used by photograph to represent the information. The Man-made environments contain many linear structures. Houses, furniture, outdoor and indoor environments in general contain a large number of planar surfaces. These surfaces often intersect each other to form line segments. This dissertation uses those line segments to find corresponding points between images. Starting from detected line segments in two or more images using phase congruency, the pairs of line segments are characterized by indices that encode the relative positions and orientations of those segments. The encoding indices are invariant with respect to viewpoint changes and the corresponding line segments between the images can be found by them. The feature matching is optimized for image synthesis where we wish to ignore unreliable matches at the expense of reducing the number of feature matches. Results are presented on real images.Generating a large number of images of an environment from closely spaced viewpoints is a very useful capability. In this paper a new method of view synthesis of the contour matching from serial images taken by camera motion based on contour information is proposed. A characteristic of the human vision is sensitive to the change of the color andluminance. The human recognizes the objects through the contours information of the objects. From this reason a new method of image synthesis based on the information of the contours is presented. The information of the contours makes it easier to be recognized and matched between two images than the other pixels in the image. This paper provides a new contour matching method between the images from the information of the relative positions and orientations of th...