Retrieving similar images in an image database using a relational matrix

The definition of perceptual similarity between two images is a function of the distance between image features, i.e, shape, color and texture features of dominant image regions. We propose a relational matrix based distance measure that can be used to measure the similarity between images in a given image database. A relational matrix, individualized for an image, describes the spatial and functional relations between neighboring image segments in the image intensity matrix. Automated image segmentation using agglomerative clustering generates the relational matrix. We propose a distance measure between corresponding relational matrices representing any two images that can be used in content based image retrieval. Simulation results are given to demonstrate the potential of the relational matrix in the digital library application.