Personalized clustering for social image search results based on integration of multiple features

Abstract

The usage of Web social image search engines has been growing at an explosive rate. Due to the ambiguity of query terms and duplicate results, a good clustering of image search results is essential to enhance user experience as well as improve retrieval performance. Existing methods that cluster images only consider the image content or textual features. This paper presents a personalized clustering method called pMFC which is based on an integration of multiple features such as visual feature, and two conceptual features(e.g., tag and title). An unified similarity distance between two images is obtained by linearly combing the three similarity measures over three feature spaces, where three weight parameters are obtained by a multi-variable regression method. To facilitate a personalized clustering process, a user preference distribution model is introduced. Comprehensive experiments are conducted to testify the effectiveness of our proposed clustering method. © Springer-Verlag 2012.