A Parallel-Serial Recognition Cone System for Perception: Some Test Results

Abstract

This paper presents some results of tests of specific "recognition cone" systems for probabilistic parallel-serial recognition and description of two-dimensional scenes of objects. Simula and Fortran encoded systems were given particular sets of transforms, and examined for their ability to hand1e scenes that contain 1)1etters, 2)place-settings of several pieces of silverware and china, and 3)a "real-world" outdoor scene, in color. The same parallel-serial flow through structured layers of variable-resolution probabilistic transforms serves to detect edges, find features, characterize, recognize and describe, without any sharp dividing lines between different types of processes. Thus, a wide variety of diverse sources of information contextually interact, in a relatively simple and general way.