A Neural Computation Model Based on nCRF and Reverse Control Mechanism - Developing Concepts in Applied Intelligence - page 45

A Neural Computation Model Based on nCRF and
Reverse Control Mechanism

Hui Wei and Xiao-Mei Wang

Department of Computer Science, Laboratory of Cognitive Model and Algorithm,
Brain Science Research Center, Fudan University, 200433 Shanghai, China
{weihui,071021059}@fudan.edu.cn

Abstract. Previous nCRF models are mainly based on fixed RF whose dynamic
characteristics are not taken into account. In this paper, we establish a
multilayer neural computation model with feedback for the basic structure of
nCRF. In our model, GC's RF can dynamically and self-adaptively adjust its
size according to stimulus properties. RF becomes smaller in local areas where
the image details need distinguishing and larger where the image information
have no obvious difference. The experimental results fully reflect the dynamic
characteristics of GC's RF. Among adjacent areas in an image, similar ones are
integrated together and represented by a larger RF, while dissimilar ones are
separated and represented by several smaller RFs. Such a biology-inspired
neural computation model is a reliable approach for image segmentation and
clustering integration.

Keywords: Non-classical receptive field, Neural network, Image

representation.

1 Introduction

Machine vision is of great significance to intelligent computing systems. However,
the vast majority of machine vision researchers overlooked the principle of visual
physiology. They proposed various algorithms from the engineering point of view.
Therefore there is a lack of unitive and effective machine vision algorithms to solve
the problem of complex scenes segmentation. In contrast, the human vision system
has a strong image processing ability. Therefore, it is entirely possible that we find
some new breakthroughs to solve the difficulties encountered in computer vision by
applying the theories of human vision, neuropsychology and cognitive psychology. In
the human vision system, the retina is also known as "peripheral brain". It processes
visual information preliminarily. The ganglion cell (GC) is the final place of retinal
information processing. The receptive field (RF) is the basic structural and functional
units for information processing in visual system. GC's RF determines the processing
result and response characteristic of the retinal complex network. Therefore, RF is
instructive for visual algorithm to simulate the mechanism of information processing
in the GC.