A NOVEL APPROACH TO THE APPLICATION OF HIGHER ORDER NEURAL NETWORKS TO IMAGE CLASSIFICATION

PDF

Authors
  1. Hatsivasiliou, F.
  2. Sala, K.L.
Corporate Authors
Communications Research Centre, Ottawa ONT (CAN);Defence Research Establishment Ottawa, Ottawa ONT (CAN)
Abstract
The applicability of higher order neural networks to the classification of low resolution imagery is investigated. A novel boundary detection and encoding methodology is developed in order to significantly reduce the large number of third order interconnection weights which must be found during the training stage of the neural network. The higher order neural classifier can be trained by presenting only one sample image per class and so enables rapid network learning with a minimal requirement for training data. An extensive, MatLab compatible toolbox developed specifically to implement this approach is described and documented along with the algorithms employed in the image boundary detection and encoding process.
Keywords
Neural networks;Boundary detection;Image classification
Report Number
CRC-96-005 — Technical Report
Date of publication
01 May 1996
Number of Pages
62
DSTKIM No
96-03428
CANDIS No
499798
Format(s):
Hardcopy;Document Image stored on Optical Disk

Permanent link

Document 1 of 1

Date modified: