ROCIT : a visual object recognition algorithm based on a rank-order coding scheme.

Gonzales, Antonio Ignacio ; Reeves, Paul C. ; Jones, John J. ; Farkas, Benjamin D.

Sandia National Laboratories

关键词: 99 General And Miscellaneous//Mathematics, Computing, And Information Science;    Pattern Recognition Systems;    Optical Pattern Recognition.;    Computer Vision.;    Computerized Simulation;   

DOI  :  10.2172/919190
RP-ID  :  SAND2004-2129
RP-ID  :  AC04-94AL85000
RP-ID  :  919190

美国|英语

来源: UNT Digital Library

【 摘 要 】

This document describes ROCIT, a neural-inspired object recognition algorithm based on a rank-order coding scheme that uses a light-weight neuron model. ROCIT coarsely simulates a subset of the human ventral visual stream from the retina through the inferior temporal cortex. It was designed to provide an extensible baseline from which to improve the fidelity of the ventral stream model and explore the engineering potential of rank order coding with respect to object recognition. This report describes the baseline algorithm, the model's neural network architecture, the theoretical basis for the approach, and reviews the history of similar implementations. Illustrative results are used to clarify algorithm details. A formal benchmark to the 1998 FERET fafc test shows above average performance, which is encouraging. The report concludes with a brief review of potential algorithmic extensions for obtaining scale and rotational invariance.