Inter-Query Relevance Learning in PicSOM for Content-Based Image Retrieval

Content-based image retrieval (CBIR) addresses the problem of finding images that the user wants from unannotated databases, based only on low-level visual features like color or texture that can be automatically derived from the images. Due to the inherently weak connection between the high-level semantic concepts that the user has in mind, and the low-level visual features that the system is using, the performance of CBIR applications often remains quite modest. One method for improving CBIR results is to try to learn the user’s preferences with learning methods such as relevance feedback. This learning is essentially intra-query, meaning that learning is started all over again in the beginning of each new query session. However, the relevance information can also be used in long-term or inter-query learning. In this paper, a method for using long-term learning in our PicSOM system is presented. It is shown that the efficiency of the system can be substantially increased by using it in parallel with MPEG-7 visual descriptors.