Spatial-textural medical image indexing based on vector quantization

This paper addresses the problem of content base image retrieval in the compression domain. We characterize the images by a representation (signature) based on low level parameters. We propose the use of spatial moments to add spatial information to the signatures based on vector quantization. The developed signatures are a combination of the histogram of occurrences of the codewords, the center of gravity and the moment of inertia associated with the histogram. The signature of each image in the base is determined, and the research is carried out by calculating, using a given metric, the distance between the signatures in the base and the signature of the query image. The methods are applied to medical databases: ultra-sound esophagus and IRM tumor brain images, and improvements of up to 18 % in retrieval efficiency are found, with regard to known usual histograms techniques.