Image Stitching Using Modified SIFT-algorithm

Computer vision is very important for the operation of technical devices such as robots, computers, automated systems targeting etc… In some applications, the image with large view becomes more and more significant. For example, the panoramic view in the photogrammetry and the panoramic video supervisory system. An approach to solve this is image stitching, which is one of the hot topics in computer vision.This research paper focuses on image stitching(mosaicking). The main processes of image stitching, which involve image registration(feature extraction and image matching), calibration, and blending, are introduced and their corresponding approaches are summarized. Then, for feature extraction, a modified SIFT(Scale-invariant feature transform) algorithm is developed to obtain better stitched image. Analysis and comparison of the image stitching method described in our thesis with the original algorithm [1, 2] of David G. Lowe are executed during the experiments. The comparison shows that the modified algorithm obtains good stitched images with less running time, using less number of keypoints.