Image Retrieval Using Circular Hidden Markov Models with a Garbage State

Shape-based image and video retrieval is an active research topic in multimedia information retrieval. It is well known that there are significant variations in shapes of the same category extracted from images and videos. In this paper, we propose to use circular hidden Markov models for shape recognition and image retrieval. In our approach, we use a garbage state to explicitly deal with shape mismatch caused by shape deformation and occlusion. We will propose a modified circular hidden Markov model (HMM) for shape-based image retrieval and then use circular HMMs with a garbage state to further improve the performance. To evaluate the proposed algorithms, we have conducted experiments using the database of the MPEG7 Core Experiments Shape-1, Part B. The experiments show that our approaches are robust to shape deformations such as shape variations and occlusion. The performance of our approaches is comparable to that of the state-of-the-art shape-based image retrieval systems in terms of accuracy and speed.