基于直方图的树与链表相结合的音频索引方法  

Audio indexing method using the combination of tree and chained list based on fuzzy histogram

作　　者：郑贵滨[1] 韩纪庆[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院

出　　处：《哈尔滨工业大学学报》2006年第11期1915-1918,1922,共5页Journal of Harbin Institute of Technology

基　　金：国家自然科学基金资助项目(60173044);教育部跨世纪优秀人才培养计划基金资助项目;哈尔滨市留学回国人员基金资助项目(2002AFLXJ005)

摘　　要：采用响度主分量模糊直方图表示音频数据,从而长度不同的两段音频数据,只要长度倍数不超过一定限度,其直方图相似度均能正确反映二者之间的包含关系.在此基础上,提出了二叉树与链表相结合的索引方法,检索时可根据检索目标的长度在索引中选择合适的层次范围搜索.实验结果表明,该索引可大幅度提高检索速度,为基于内容的音频信息检索提供了一种有效的索引方法.A kind of fuzzy histogram based on principal loudness is firstly proposed, so that when the length ratio of two audio segments is not bigger than an upper limit, the similarity between their histograms can correctly reflect that the longer audio segment contains the other one. Therefore, investigators propose a novel au- dio indexing method using the combination of binary tree and chained list. When retrieval, different index level is selected for searching according to the length of target and the upper limit. Experimental results show that the retrieval speed can be effectively improved by the index, especially when the target is longer. The method is an effective solution to content based audio information retrieval.

关 键 词：音频检索 索引 直方图 二叉树 链表 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]