A Novel Image Retrieval Method Based on Mutual Information Descriptors

摘要

In this paper, we propose a novel image retrieval method based on mutual information descriptors (MIDs). Under the physiological property of human eyes and human visual perception theory, MIDs are extracted to encode the internal correlation relationship among multiple image feature spaces, characterizing image contents with mutual information features based on the low-level image features, such as color, shape etc., then, the mutual information features fusion strategy is used to imitate the information transfer process in nervous system. When using the MIDs proposed to image retrieval, we can get many advantages such as low dimensionality, a certain robustness of geometric distortions and noise, and describing the human visual retrieval mechanism effectively. Experimental results show that MIDs have high indexing and retrieving performance compared with existing methods for content-based image retrieval (CBIR).