Research On The Key Algorithms In Panorama Constration

Panorama is an image with a wide-angle view. It can be taken with special cameras or be constructed by image alignment techniques from photos taken by regular hand-held cameras. The techniques to create panoramas are among the hottest topics in the field of IBR (Image Based Rendering). Panorama recognition, also called image stitching or image mosaic, has been extensively studied for a long time because of its variety and complexity. It has applications in virtual reality, satellite image and medical image processing.Based on analysis of the existing algorithms, this thesis summarizes the key algorithms implemented in the panorama construction research. First, because of shortcomings of traditional template matching, this thesis proposes a new adaptive template matching algorithm. The reference template will be constructed in the feature object area with location determined by local edge density and a fuzzy clustering algorithm. The template is then used to compute its local entropy difference and to search in another image to find the best matching point. Second, we study a feature-based stitching algorithm. We choose a new multi-scale Harris corner algorithm to select feature points, then use a robust feature point matching algorithm to determine the image transformation parameters. Third, to solve ghosting and abruptness in stitched images, an algorithm based on dynamic programming is used to find the best seam-line to stitch the images. Experiments show that the above-mentioned algorithms can implement accurate matching of images and eliminate seams and ghosting in stitched images.Finally, we developed an automatic panorama construction system using Microsoft Visual C++6.0.