Image Key Point Detection Algorithm Research And Image Stitching Application

Image local invariant features are very widely used today. We do some researches on the keypoints detectors and image mosaic based on image local invariant features in this dissertation.Firstly, this dissertation expounds image local invariant features, and puts forward a new detection algorithm which can reduce the testing time and lower the false corners through making non-maxima suppression within the circular region. At the same time, the new algorithm can significantly reduce the differences of detection in image rotation without changing any parameters. Experimental results confirm the validity of the new method and show that the new method greatly improves image corner detection performance with higher accuracy compared with the traditional algorithm.Secondly, based on the multi scale-space of image, this dissertation explains the multi scale structure of Pyramid model and Gauss model. And then it analyzes the SIFT algorithm based on these models, by the way, describes the application of multi scale-space structure in image keypoints detection in detail. Based on the analysis of the classical SIFT algorithm, this paper introduces an improved SIFT algorithm which mainly ameliorates the algorithm on keypoints descriptors in order to better describe the detected image keypoints and make it play a better effect on the subsequent application.Finally, an improved algorithm for image mosaic is proposed based on an improved Harris corner detection. We calculate the similarity of the target pixel within 8 neighborhoods and make non-maxima suppression within the circular region to improve the traditional algorithm. Simulation results show that this algorithm increases the accuracy, stability and rapidity of image mosaic. In a word, this algorithm can reach a target of seamless mosaic and has good practicability.