Fast Orthogonal Neural Network for Rotation-Translation- and Scale-Invariant Image Recognition

B. Stasiak (Poland)

Keywords

Fast Orthogonal Neural Networks, Image Recognition, Signal and Image Processing

Abstract

In this article a novel method of image recognition invari ant under rotation, translation and scaling is presented. The proposed method is based on a fast orthogonal neural net work which, due to its structural analogy to fast algorithm for Fourier amplitude spectrum computation, enables to classify images irrespectively of their translations. This property, in conjunction with log-polar representation of image amplitude spectrum, is subsequently applied to ob tain also the rotation and scale invariance. The proposed classi´Čüer is compared to a multilayer perceptron and to k-nearest neighbors method, showing its superiority in a series of tests performed on specially constructed image databases.

Important Links:



Go Back