Title: |
Multi-Resolution Simple Cell Model |
Source: |
Proceedings of 2010 Cross-Strait Conference on Information Science and Technology (CSCIST 2010 E-BOOK)
(pp 662-666)
|
Author(s): |
Yaping Huang, Dept. of Computer Science and Technology, Beijing Jiaotong University, Beijing Qi Zou, Dept. of Computer Science and Technology, Beijing Jiaotong University, Beijing Mei Tian, Dept. of Computer Science and Technology, Beijing Jiaotong University, Beijing |
Abstract: |
This paper constructs theoretical models motivated by neural processing a novel Multi-Resolution Simple Cell Model, namely MRSCM, which attempts to provide a framework for understanding the function of neurons in retina, LGN and V1. It consists of three processing steps: firstly, multi-resolution images are obtained by using Gabor filters to analyze the properties of single neuron, and then linear transformation ICA is adopted based on higher-order properties of the environment, finally, the basis functions at the each location and each resolution are constructed as a subspace, and its projection distance is estimated as the output of the models. MRSCM accords with the multi-resolution neural processing mechanism and the sparse coding features, and also leads to simultaneous emergence of both oriented and bandpass properties similar to visual cortex simple cell.
|