Audio Fingerprint Oriented Quantum Hash Technology Research

With the advancement of various software and hardware tools to copy, distribute and generate multimedia data, the number of multimedia data on the web has grown in lips and bounds. How to robustly and efficiently retrieval the interesting information form the internet is urgently to resolve in recent years, content-based digital audio fingerprint technology is a good choices since it allow the independence of the audio format and don't have to embed any other surplus information, small data amount and good robustness. It can do the automated retrieval and the recognition of the audio data, and have a fast development in recent years. When the audio under the experience of distortion especially the noise contaminate, the performance of the system will fall rapidly, so efficiently and robust audio fingerprint is the pursuing goal of many academician and organization research on it. Current paper have expanded the intensive study of the audio fingerprint technology, and made some impressive achievements that are as follows.Propose a novel robust audio fingerprint oriented quantum hash , base on the difference of the subband energy as the audio feature, encode the distortion between the query audio and the candidate audio into a integer as it's quantum distance according to its probability density function, it can better depict the distribution of the distortion between the query and the candidate, and ameliorate the situation that the binary procedure that many traditional audio fingerprint system based converts a real value to a binary value will cause the information lossy which can reduce the robustness of the audio fingerprint system.Propose a two-stage match scheme in allusion to the audio quantum hash fingerprint, the first stage are based on LSH which discard the fingerprints that impossible in the database, and the left compose the candidate database ; the second-stage adopt the typical matching method, and the quantum hash distance as the similarity metric, matched with the candidate from the candidate database one by one to find the results that satisfy the specific precision requirements of the user. Experimental results show that the proposed method is more robust under various distortions than the binary hash.