Development tool for computer vision systems at pixel level

Abstract

This paper describes a tool to aid the development of computer vision applications for pixel-level diagnosis. A conceptual framework for constructing pixel processors is described. The proposed tool is based on two domains: one deals with numerical images, named features, and the other deals with symbolic images, named classes. A reduced set of image and operator types is defined. Based on it, an object-oriented language is described. The main objects define features and classes and their associated operators, namely procedures, classifiers, and rules. All system control is goal oriented and is designed to provide symbolic images as results when a request is received from a higher level of the image understanding system. Some knowledge engineering concepts, such as certainty control and instances, are also included.