Integrating Textual Semantic And Visual Content For Web Personal Image Retrieval

Along with the development of network and multimedia technique, it becomes an urgent problem how users can get the exact information they need quickly among the mass multimedia data on Internet. As a prominent form of multimedia, image retrieval has become an important project research presently. Nowadays, the development of image search engine mainly bases on two kinds of technique: 1) traditional Text-Based Image Retrieval (TBIR); 2) Content-Based Image Retrieval (CBIR). However, because of the limitation from the "Semantic Gap", the two both have their drawbacks. To get the best out of their advantages, this dissertation explores the combination of the web textual semantic mining and image visual content judgment deeply and builds a prototype system for web personal image retrieval.The main innovative work of this dissertation is represented below:(a) Propose a system framework for the web personal image retrieval on Internet, which integrates the textual semantic mining and visual content judgment.(b) Realize the judgment of the visual content of web personal images using AdaBoost face detection technique; at the same time, fill the results returned from large commercial image search engine via extracting the correlated textual semantic information embedded in web document.(c) Propose a model for the web textual semantic mining based on PLSA(Probability Latent Semantic Analysis). And based on this model, carry out an adapting dynamic weighting schema on different information elements.(d) Build a prototype system for web personal image retrieval on Internet, which comprised of network crawler module, visual content judgment module, textual semantic mining module and server interface module.