Labeling schemata and syntactic descriptions of pictures

Abstract

Advocating the view that the so-called pattern recognition problem is much more appropriately studied as a problem of pattern analysis and description, a specific syntactic descriptive model is proposed for classes of pictures composed of linelike elements. The implementation of this model envisages a two-stage processing of digitized input pictures. A formalism is worked out in detail within which the labeling algorithms used in the first stage can be efficiently described and studied. A system structure for a computer to carry out this class of picture processing operations-called parallel processing-is outlined and it is shown that a natural extension of ALGOL can be made to serve as an adequate programming language for such a computer. The extension of this model and the processing details to deal effectively with noisy input pictures is considered. Some criteria for acceptable noise-cleaning techniques are discussed. Several examples of labeled pictures are included to illustrate the validity and power of the proposed model and the processing techniques.