Combining Morphological Filters And Deformable Models To Design A 2-D Shape Based Retrieval System

In this paper we present a Content-Based Images Retrieval (CBIR) system which uses characteristics of a region occupied by a shape and the outer contours of the same. The CBIR system is a two-stage image retrieval system: it first extracts a subset of images which contains those shapes most similar to those contained in the query shape, based on the use of morphological characteristics of the objects considered. It then indexes an extracted subset of shapes from the database in view of the similarity of the deformation of the outer contours of these shapes. For this first stage, we propose a geometric method of two-dimensional shape recognition based on the use of characterizations of those curves given by the morphological operators of erosion, closure, aperture and dilation with structural elements of increasing size. The second stage of this system uses an invariant characterization of shapes based on a deformable model. In the experiments, we have illustrated that the CBIR system can accurately retrieve the most similar images from the database by using scaled, rotated, small deformed or mirrored sketched query images