IMAGE CLASSIFICATION BY NEURAL NETWORKS USING MOMENT INVARIANT FEATURE VECTORS

PDF

Authors
  1. Sala, K.L.
Corporate Authors
Communications Research Centre, Ottawa ONT (CAN);Defence Research Establishment Ottawa, Ottawa ONT (CAN)
Abstract
As image classification system based upon the extraction of moment invariant feature vectors and an artificial neural network classifier is described. The moment invariant feature vectors are derived from the test images using series of orthogonal basis functions. Six different basis functions are studied which include four types of Zernike functions and two types of Walsh functions. Four different schemes for the normalization of the feature vectors are also investigated. The images used in the study possess random scales, lateral positions, and angles of orientation in the image plane. In addition, random noise with different signal-to-noise ratios is superimposed upon the images. The feature vector extraction technique employs the concept of moment invariants so that the feature vector components are independent of the image's scale, lateral position, and orientation. The neural network employed for the classification task is a multilayer perception network which is trained with the back propagation algorithm. The performance of the overall classification system is determined by measuring the classification accuracy as a function of the signal-to-noise ratio of the test imagery. The work and the results presented in this study form the basis for a neural network based, image recognition system which will be employed in the classification of military, synthetic aperture radar (SAR) imagery of land targets.
Keywords
Neural networks;Moment Invariant Feature Vectors;Image classification;SAR imagery
Report Number
CRC-97-002 — Technical Report
Date of publication
01 Feb 1997
Number of Pages
259
DSTKIM No
97-02702
CANDIS No
502966
Format(s):
Hardcopy;Document Image stored on Optical Disk

Permanent link

Document 1 of 1

Date modified: