Invariant to transformations image retrieval from image databases using a boundary based description of object shape

Abstract

The basic problem of the methods for shape description of image objects is their dependence on the graphical transformations. In this paper we give boundary based method for shape description based on a multi-step criterion. We present a new approach for similarity shape retrieval from image database that achieves invariance of results with respect to transformations. We define and use a criterion to achieve invariant image object representation with respect to rotation. The approach includes original description of image objects shape of the type of multi-metric attribute stored in image database and a similarity distance definition for image retrieval processing that is invariant to reflection too. Our approach achieves results that are invariant with respect to arbitrary compositions of graphical transformations.