Research On Rotational Panoramic Image Stitching Algorithm

Panoramic image stitching technology solves the problem of obtaining a high resolution panoramic image through low-resolution images with overlapping regions of the same scene. The main process contains feature detection and image registration.Feature detection is a problem of extracting key points,which is the premise of image registration. Image registration aims to find optimal homography to map image coordinate to another one. In this paper, we carry out research on feature detection and image registration. In the feature detection stage, it was found that redundant image features detected by SURF algorithm, which cause much time in the generation of feature descriptor and image registration. Thus, we present an improved algorithm based on BP neural network to predict the appropriate number of features. In the image registration stage, we first use a KNN algorithm for rough feature matching, and then use the RANSAC algorithm for the purification of the feature points to get optimal homography. Finally, the stitched image is generated by the weighted average method which achieved good results.The research contents are summarized as follows:1. Presenting a method for removing false features matching based on KNN and RANSAC. We select the SURF algorithm as the basis of image registration, and acquire panoramic image by weighted average method.2. Presenting two methods to accelerate image stitching time. One method for reducing the search image range is implemented by using an improved NCC algorithm to lock overlapping regions and another method by SURF threshold estimation based on neural network to reduce the number of feature points. Experiment results show that the proposed algorithm can accelerate image stitching time with good result.3. The rotary image stitching software was built based on MFC, which verified the efficiency of the improved algorithm in stitching time and imaging effect.