Shape Retrieval Based on the Relativity of Chain Codes

Abstract

The direction chain code has been widely used in image retrieval for its simplicity and low storage requirement. One problem is that the traditional methods did not consider the relativity among the chain codes, which limited its use in image retrieval. In this paper a novel shape feature called chain code relativity entropy (CCRE) is proposed for shape classification and retrieval. The direction chain code is firstly mapped to different state of a Markov chain and a new transition probability matrix is introduced. Then, relativity histogram is defined, which includes the transition probability of one state to the others and the others to this one. Based on relativity histogram and information theory, we give the definition of CCRE. After that, the characters of CCRE and an improved method are discussed. Comparisons are conducted between the CCRE and several other feature descriptors. The results show that the CCRE is efficient and it provides noticeable improvement to the performance of shape retrieval.