Improved SIFT Algorithm And Its Application In Image Mosaic

Image mosaic technology has become a research focus in the image processing, it is to two or more pieces of overlapped photoes by the method of using a certain space matching, after sampling the fusion to form a picture contains various image sequence information of wide field of vision, complete, high-definition seamless high-resolution images. Through researching of existing SIFT algorithm, an improved algorithm is proposed in view of the insufficiency in the existing algorithms. The work includes the following three aspects:(1)Since Scale Invariant Feature Transform(SIFT) feature point extraction and feature descriptor incur large amount of calculation, the improved SIFT feature points extraction algorithm is proposed.The algorithm use gaussian second-order difference operator of zero point detection to extract the SIFT feature points, simplifies the structure of the pyramid, and the characteristics of the phase, through the use of round rotation invariance, puts forward an improved SIFT feature descriptor, with circle to construct characteristic descriptors, avoid the distribution of the main direction and a lot of calculation, shortening the time of character description, improves the rapidity of the algorithm.(2)Since the SIFT algorithm is applied problems of long time i n the process of feature matching and wide range of searching, a n ew matching method of SIFT based on block area is proposed. The method splits image registration into bilateral block, and the impro ved SIFT is used to extract feature points of block matching area a nd using block matching strategy for match elementarily,with dispari ty gradient constraints to eliminate the false matching. The experim ental results show that improved SIFT algorithm of area blocking g uarantee the accuracy of registration at the same time reduce the m atching time, narrow the scope of matching and improve the matchi ng efficiency.(3)Existing weighted smoothing methods causes obvious seam-lin e due to large intensity difference between the two images. A linea r weighted smoothing algorithm for the optimal stitching-line remov al is proposed. The proposed algorithm is based on the observation that small gradient and divergence indicate small intensity variation and gradient direction variation respectively. The optimal stitching-l ine is selected as the minimum sum of the gradient and divergence of 3*3 local patch in the overlapped image region.Weighted smoot hing algorithm is used to remove stitching line. Experiment result s hows that seamless stitching can be achieved by the proposed meth od.