Abstract :
In this thesis, four typical algorithms for image feature extraction are studied first, by analyzing their advantages and disadvantages and according to the actual needs of the project, SIFT algorithm is selected to be the foundation of the research. By analyzing the disadvantages of SIFT and its relational improved model, based on the two way guiding and matching strategy, have proposed a new matching strategy with RANSAC and LM, with the filter of feature match points and the search of inliers, has solved the problems of big match error and low accuracy in the original SIFT algorithm. According to the sub-tests for improved algorithm in every period of image mosaic, the results show that the improved algorithm is better than the original one. In addition, the universality of improved algorithm has been tested, the result shows that the improved algorithm can meet the needs of various types of image mosaic, the universality of the algorithm is strong.
Keywords :
feature extraction; history; image matching; image segmentation; iterative methods; transforms; LM; Levenberg-Marquardt algorithm; RANSAC; SIFT algorithm; feature match point filter; frescos; guiding strategy; image feature extraction algorithm; image mosaicing algorithm; inlier search; match error; matching strategy; random sample consensus algorithm; scale invariant feature transform algorithm; Accuracy; Algorithm design and analysis; Feature extraction; Image edge detection; Image fusion; Purification; Vectors; Digital image processing; Image Mosaicing; SIFT;
