Research On Automatic Annotation Based On Image Saliency And PLSA-GMM

Because of the “semantic gap”,automatic image annotation(AIA)is getting more and more attention in the field of content-based image retrieval.In the paper,we firstly give a briefly introduction of the state and the key technology of AIA.Then,the new schemes are introduced based on GMM(Gaussian mixture model),PLSA(Probabilistic latent semantics)and feedback log for higher annotation performance.The main work and innovation of this paper are as follows:(1)The key technologies of AIA are analyzed and summarized in detail,which include image low-level feature,the model of AIA,the feedback technology,the similarity measurement and the performance evaluation of the AIA methods.(2)In order to reduce the influence of image background,a new annotation method is proposed based on the idea of the division of the image foreground and background.Firstly,the low-level features of the background and foreground are extracted respectively.Then,an implicit subject is inserted between the low-level features and the image labels by PLSA to build the relationship of the subject and the annotation word.Finally,the annotation scheme is built by GMM to determine the parameters of the model.Further,another annotation method is proposed based on the multi-features(color,shape and texture)and PLSA-GMM.(3)In order to reduce the “gap” between the low-level features and the image semantics,the feedback log technology is introduced into the AIA.Based on the feedback log,WordNet and PLSA-GMM model,the performance of the above schemes can be improved greatly for image annotation.(4)Based on the above annotation models and the Java language,we design a prototype system for image annotation and retrieval.