Describing objects by a multi-resolution syntactic approach

Abstract

This paper describes, in a multiresolution environment, a method for 2-D object recognition which relies on a linguistic description of the object contour at different resolution levels. The object is firstly digitized at the maximum resolution level and then, by using a special smoothing technique for avoiding the addition of noise, is represented at different resolution levels until the coarsest level. A parallel, context sensitive grammar (Kirsch-like), is then used for defining production rules which describe the evolution of the contour description (following the production rule) between levels. In order to recognize the object, a matching must be performed between its string descriptions and those of prototype objects at each corresponding resolution level. The main advantages of this approach are computational efficiency (due to fast search on a coarse representation guiding the detection of contour segments on levels at higher definition) and rotation independence as well as noise immunity.

This work has been partially supported by the Italian National Research Council. Progetto Finalizzato “Siatemi Informatici e Calcolo Parallelo”, Sottoprogetto Calcolo Scientifico per Grandi Sistemi.