The Research Of Cross-Media Retrieval Technology

In the fast developing society of information and networking, cross-media content hasbecome an important research focus in the information retrieval comprehend. The type withtraditional information retrieval, the cross-media information retrieval object is not a singletype of multimedia object, often including images, sounds and video format data. In thestudy of cross-media field of information retrieval, for the different types of multimediainformation retrieval is not yet mature stage. People the identification process information,often need to carry out the different types of information perceptual comprehensive analysis,including visual, auditory, and other information, is formed integrally cognition.For the development of cross-media retrieval technology, is inseparable from thetraditional content-based multimedia retrieval technology. Since the1990s, a variety ofcontent-based multimedia retrieval technology of computer vision, pattern recognition,database technology, and machine learning technology together, to form a multi-angleanalysis of the mechanism, to make up for the traditional single type of multimediainformation retrieval drawbacks, greatly improving the efficiency of information retrieval inthe case of large amount of data. Cross-media information retrieval initial content-basedretrieval techniques, used to extract a variety of information of the image, such as color,texture, and shape feature vector as the entry point of the image index. Based on thistechnology, can be transferred to a content-based retrieval of video and audio data, stillmore satisfactory results can be achieved. However, these methods are mostly based on asingle type of multimedia data as a search object cannot be achieved between different typesof data often cross retrieval, e.g. the use of the audio data with the image data comparison,and the face with the dual voice recognition problem can be achieved.The underlying thesis of the media feature extraction were studied in detail, and toreduce the use of PCA dimension calculation. For audio examples, using the MPEGcompression of audio data, the use of fuzzy C-means clustering algorithm for clusteringanalysis extracted centroid, with Mel cepstral comparisons and calculations, the similarityof the results; For images using wavelet transform to extract image edges, colors in theimage analysis feature, texture characteristics, by calculating the composition of its seven invariant moments feature vectors expressing this image to calculate the similarity distancefunction to get the result.This study is based on a single type of multimedia data retrieval technology,cross-media data retrieval. Cross for image, audio and other multimedia data, the minimumsize of the data underlying the characteristics of victimization, with the convergence of thecorrelation matrix operations, cross-media analysis of image and audio data, and throughmultimedia data drop dimension, and maintain a persistent correlation learning.