A Refined Hybrid Image Retrieval System using Text and Color

Abstract

Image retrieval (IR) continues to be most exciting and fastest growing research areas due to significant progress in data storage and image acquisition techniques. Broadly, Image Retrieval can be Text based or Content based. Text-based Image Retrieval (TBIR) is proficient in „named-entity ‟ queries (e.g. searching images of „TajMahal‟. Content Based Image Retrieval (CBIR) shows its proficiency in querying by visual content. Both the techniques having their own advantages and disadvantages and still have not been very successful in uncovering the hidden meanings/semantics of the image. In this paper, we propose a hybrid approach that improves the quality of image retrieval and overcomes the limitations of individual approaches. For text retrieval, matching term frequency-inverse document frequency (tf-idf) weightings and cosine similarity are used, whereas for content matching the search space is narrowed down using color moments and then the two results obtained are combined to show better results than the individual approaches. Further refinement using color histogram technique improves the performance of the system significantly.