文档介绍：Abstract
In recent years, mass multimedia data appears online. A large number of repetitive data in the database not only causes a huge waste of storage space, but also brings difficulty to information retrieval and query. Therefore, research of duplicate detection for multimedia data has important practical significance. This paper tries to give an answer about how to achieve multimedia data duplicate detection in large-scale databases. Its main content is as follows:
The Philips algorithm shows a good robustness under most signal distortions, but the results are not satisfactory in the real noisy environment. The MBM (Bit Mask) algorithm has a better noise pared with the Philips algorithm in the actual environment. However, the MBM algorithm’s retrieval efficiency and robustness will get worse with the expansion of the database size. In this paper, we proposed a novel audio fingerprint extraction based on harmonic filter according to the advantages and disadvantages of the two above algorithms. Experimental results show that the audio fingerprints extracted by the new algorithm have better noise robustness than the Philips algorithm and have higher retrieval efficiency than the MBM algorithm. In addition, we use multi-stage retrieval method to improve the retrieval precision further.
Because large-scale multimedia database have very large amount of audio fingerprint data, duplicate detection faces the problem of memory shortage and low detection efficiency. In this paper, segmentation-based data loading method is proposed. Based on this method, a copy detection method is given in detail and a method to detect duplicate of sub-series of multimedia files is designed. Experimental results show that the duplicate detection method proposed is effective.
Based on the research of audio fingerprint scheme and duplicate detection method, we design a multimedia database duplicate detection system. Before the multimedia files are added into the database, the pleted dupli