Research On Hash Algorithm Based Image Copy Detection

As the development of computer internet and technology of multimedia compress, thedistribution of multimedia become more and more convenient, makes the distribution of thecopy of multimedia easier. So, how to protect the copyright of digital media is the majorproblem at now. As one of the methods of copyright protection, content-based copydetection technology includes two parts: feature extraction and index construct.Based on reading the literature at home and abroad, we select the global information asthe feature of image. We can use a lot of classic algorithm to process the image matchingbecause one image correspond one point in high dimension space and the feature iscompressed. For GIST features, we transform the original feature vector into binary vectoruse Principal Component Analysis, Ordinal Measure, Locality Preserve Projection andMaximum Entropy Model. Because of ordinal measure is based on the order sequence offeature vector and locality preserve projection can keep the local structure information oforiginal data, so after processing, binary vector has lower dimension and can resist somegeometric attacks.The other problem of copy detection is index construction. To high dimensional featurevector, the traditional index construct algorithms face the problem of “dimensional disaster”,.In this paper, we propose a new index structure based on binary code. The same codes willbe stored in the first index and the second index store the similar code to the current code. Inthe search progress, we use recursive method to search new set of code in second index.From experiments we find that our feature extraction algorithm has better robustnessand the search time of our index structure is shorter than LSH. For a lot of query images, wepropose a search algorithm based on group method. Compared to random search, our methodhas less IO counts and improves the efficiency of our system.