Representing pattern recognition-embedded systems through object-process diagrams: the case of the machine drawing understanding system

Abstract

Pattern recognition involves a host of problems, algorithms and techniques dealing with all aspects of detection and classification of objects in scenes and their conversion into meaningful interpretations. Frequently, such tasks are embedded within vision systems that are themselves embedded within more complex systems. Concentrating on minute, albeit important, details of some pattern recognition algorithm, may result in blurring of the “big picture” that puts the algorithm under consideration in the more general framework. Pattern recognition (PR)-embedded systems feature a combination of complexity on the one hand and a balance between structure and behavior on the other hand. Analysis and understanding of such systems call therefore for a methodology that represents equally well structure and behavior within a unified frame of reference and has adequate tools for complexity management. This work proposes the object-process analysis (OPA) as an approach to tackle this task. The Machine Drawing Understanding System (MDUS) is as an instance of a PR-embedded system used as a case in point. We provide a motivation for the development of the system in general and its specialized Orthogonal Zig-Zag (OZZ) vectorization algorithm in particular. To demonstrate the suitability of OPA for representing PR-embedded systems at any level of detail, we apply it to communicate a top-down introduction of MDUS and its OZZ algorithm. The result is a series of consistent, inter-related object-process diagrams that gradually expose the details of the system. Complexity is managed through visibility control, which is obtained by a host of options for scaling object process diagrams. The ease of application of object-process analysis to the case in point suggests that it can be successfully applied to analyze, understand and communicate PR-embedded systems.