Perceptual metrics for image database navigation

Yossi Rubner, Carlo Tomasi0792372190, 9780792372196

With the increasing number of images available electronically, automatic retrieval systems are becoming essential. This book introduces an absolute prerequisite for any such system: a metric, called the Earth Mover’s Distance (EMD), for comparing images in terms of their appearance. This metric describes the amount of work that is necessary to transform one image into another, in a precisely defined mathematical sense, and in a flexible and perceptually meaningful manner. An efficient linear programming algorithm enables the computation of this metric fast enough to be used for the interactive retrieval of images from large repositories. The perceptual properties of the EMD, and the speed of its computation, lead to database navigation, a new paradigm for interacting with a repository of images. When navigating, the user is shown a very large number of images in response to a query. The EMD between pairs of images, together with a multidimensional scaling method, allows these images to be displayed so that similar images appear near to each other on the computer screen. In this way, the user can grasp at a glance what is returned, and can reach the images of interest with a small number of mouse clicks. A CD-ROM with full color images is included.
Extensive benchmark evaluations and example retrieval systems show the usefulness of the EMD and the advantages of image database navigation.
This book will be of interest to researchers, industrial professionals, and graduate and post-graduate students in the fields of Computer Vision; Image Processing; Data Mining; Digital Libraries; Psychophysics; Computer Science; Electrical Engineering.