Research On Duplicate Detection Based On Audio Fingerprint For Multimedia Database

In recent years, mass multimedia data appears online. A large number ofrepetitive data in the database not only causes a huge waste of storage space, butalso brings difficulty to information retrieval and query. Therefore, research ofduplicate detection for multimedia data has important practical significance. Thispaper tries to give an answer about how to achieve multimedia data duplicatedetection in large-scale databases. Its main content is as follows:(1) The Philips algorithm shows a good robustness under most signaldistortions, but the results are not satisfactory in the real noisy environment. TheMBM (Bit Mask) algorithm has a better noise robustness compared with thePhilips algorithm in the actual environment. However, the MBM algorithm’sretrieval efficiency and robustness will get worse with the expansion of thedatabase size. In this paper, we proposed a novel audio fingerprint extractionbased on harmonic filter according to the advantages and disadvantages of the twoabove algorithms. Experimental results show that the audio fingerprints extractedby the new algorithm have better noise robustness than the Philips algorithm andhave higher retrieval efficiency than the MBM algorithm. In addition, we usemulti-stage retrieval method to improve the retrieval precision further.(2) Because large-scale multimedia database have very large amount of audiofingerprint data, duplicate detection faces the problem of memory shortage andlow detection efficiency. In this paper, segmentation-based data loading method isproposed. Based on this method, a copy detection method is given in detail and amethod to detect duplicate of sub-series of multimedia files is designed.Experimental results show that the duplicate detection method proposed iseffective.(3) Based on the research of audio fingerprint scheme and duplicate detectionmethod, we design a multimedia database duplicate detection system. Before themultimedia files are added into the database, the system completed duplicatedetection on these multimedia files. Considering the huge computation load underlarge-scale databases and network platform, a distributed structure is used in thedesign of detecting system. The experimental results show that the systemachieved the expected goal of duplicate detection.