Relevance feedback technique for content-based image retrieval using neural network learning

Relevance feedback (RF) is an interactive process in content-based image retrieval (CBIR), which refines the retrievals to a particular query by using user's feedback on previously retrieved results. In this paper, by changing the process of relevance feedback into a learning problem of neural network, a relevance feedback technique for content-based images retrieval by neural network learning (NELIR) is introduced, which can improve user interaction with image retrieval systems by fully exploiting similarity information. NELIR can describe the distribution of positive feedback sample images in feature space with a set of neighboring clusters produced through constructing neural network, for accurately reflecting their semantic relevance. In particular, constructing neural network is dynamic. The neural network depends on which images are retrieved in response to the query. On the other hand, NELIR is independent of the specific feature extraction and similarity measure. Thus, it may be embedded in many current CBIR systems to improve the performance of image retrieval. The performance of a prototype system using NELIR is evaluated on a database of 2,000 images. Experimental results demonstrate improved performance compared with a traditional CBIR system without NELIR algorithm using the same image similarity measure.