Abstract
Since intra-session learning in content-base image retrieval (CBIR) system infers user’s preference according to merely the information of current relevance feedback session, many researchers now attempt to accumulate and utilize the knowledge obtained from previous query sessions. This thesis presents a method to infer hidden semantic cues by accumulating the knowledge learned from relevance feedback sessions. We propose to estimate the explicit relations between hidden semantic concepts and images using a probabilistic model. In short-term learning, we apply the general SVM classification to initialize the hidden semantic space. Once the accumulated hidden semantic space becomes impractically large, we propose using support vector clustering (SVC) to construct a more compact and still meaningful hidden semantic space with lower dimensionality. Given a dimension-reduced hidden semantic space, we then perform the image query in terms of the hidden semantic attributes instead of merely the visual features. Our experimental results and comparisons demonstrate that the proposed hidden semantic feature representation as well as the SVC-based technique indeed achieves promising results.