| With the rapid development of image acquisition, storage and network technology, the images have gradually become an indispensable part of people's life, for they have been widely used in many areas. As the image-processing software has been continuously enriched and improved, a lot of negative impacts will be inevitably brought to our life because the images can be easily tampered. The authenticity and integrity of the images are faced with enormous threats and challenges, therefore to study the detection methods of image tamper is necessary.In this paper, we improved the SIFT matching algorithm, and study the image copy-paste forgery detection based on SIFT and the improvement. The main contents are as follows:(1) Through a large number of literature and experimental comparison, we find that that the presence of matching accuracy and time complexity in the algorithm based on SIFT are high. In this paper, a new type of feature extracting and feature matching methods are proposed to deal with these questions.(2) The SIFT algorithm has been improved in this paper: the threshold value is determined by the method of fitting optimization to solve the problem of inaccuracy when the threshold is increased, and the methods of extracting the feature points by SIFT algorithm are improved; the BBF search algorithm based on K-d tree is adopted to achieve the fast matching of the nearest neighbor query, and the feature matching is improved in SIFT algorithm, so the problem of high time complexity goes easy.(3) Based on the above work, a further improvement has been made in this paper for the high feature vector dimension. To locate the main direction of the feature points and feature point descriptor by Haar wavelet transformation, then to reduce the dimensionality of each feature point by using the PCA algorithm, and the computation complexity will be reduced,the calculation speed of the algorithm will be improved.Experiment results verified the effectiveness of the algorithms of our paper. |
PDF Full Text Request