Pictorial Editing by Shape Matching Techniques

Abstract

A number ofpainting and retouching packages operate much like the artist s traditional canvas andpalette, with electronic toolsfunctioning like their studio counterparts. These programs, while offering an ample set of tools for creating an image, suffer from an intrinsic limitation as regards its modification, which lies in the fact that changes to the shape of objects and those requiring some kind ofpattern recognition are generally difficult and cannot be accomplished automatically. This paper deals with an original methodfor providing a pictorial editor with the"search-and-replace" facility, that works rather like text substitution in a word processor. The user defines the search (model) and the replacement (target) patterns by example, i.e. by showing the system the patterns taken from the image or an existing catalogue. The editor then searchesfor objects that match the model s pattern and replaces them with the target one in an automatic or user-controlled mode. The method is based on a model-driven matching technique, capable of measuring the similarity of objects that are partially occluded or transformed by translation, rotation, or change of scale. Salient features of the method are its robustness and the limited number of parameters needed to adapt the search procedure to different application contexts. A search-and-replace function used for pattern-editing in the field of textile design, is presented as a working example.