A novel sift based algorithm for image stitching in panoramic images

A wide-angle picture, which is a combination of multiple images including common area, is called a panoramic image. Panoramic image is used in many applications such as mouth, tooth and jaw radiology, geodesy and photogrammetry, virtual navigation. When obtaining panoramic images, the key points of the pictures to be merged are firstly subtracted and the common feature points are determined. The wrong features matched in the two images are eliminated and the images are merged with the registration process. In this thesis, a novel hybrid method based on SIFT is proposed for image stitching process. The union set of points obtained from the SIFT and Harris methods was used to obtain stronger points in feature extraction. In eliminating the wrongly-matched points with more points, the images are merged by using the ratios calculated from the radiuses of the circumference circles obtained from these points. To achieve this process in a shorter time, different feature extraction methods were applied on images in parallel. In the experiments, it was shown that SIFT based hybrid method supported by sinus theorem was a more robust approach than the current recursive and threshold-based matching elimination methods.