CROSS-MEDIA RETRIEVAL METHOD BASED ON GENERALIZED LINEAR REGRESSION MODEL

摘要

Disclosed is a cross-media retrieval method based on a generalized linear regression model. The method, the semantic features of different mode objects are extracted first, then the generalized linear regression mode is utilized to establish regression relations among mode features to implement mutual conversion of the different mode features, subsequently, several types of logistic regression algorithms are utilized to estimate a posterior probability distribution of the converted mode objects, and finally, a distance-measuring method is utilized to calculate the distance between a test sample and a database sample, thus outputting the first N-number of most similar samples of the database that are acquired by retrieval. When crossing a gap between the semantics of the different modes, the present invention is capable of preventing as much as possible the disclosure of effective information when converting between the different modes of media, thus ensuring effectiveness in transmitting the different mode information and further improving the robustness and accuracy of cross-media searches, and providing a great application prospect and a significant market value.