User-adaptive image clustering using relevance feedback for efficient content-based retrieval

In content-based image retrieval (CBIR), similarity measures vary according to the user, and it is difficult to build a retrieval system which reflects the user´s similarity measures automatically. Regarding CBIR as consisting of feature extraction, coarse classification and detailed matching stages, this work aims at reflecting the user´s similarity measures in coarse classification. After obtaining the user´s evaluation to the initial retrieval, we transform the initial feature vectors using optimal linear associative memory (OLAM). This leads to the selection of important features from the user´s relevance feedback. Experimental results show the effectiveness of the proposed method which reflects the user´s similarity measures in the coarse classification.