Fourier descriptors based hand gesture recognition using neural networks

Abstract

With the advanced computing and efficient memory utilization, vision-based learning models are being developed on a large scale. Sign language recognition is one such application where the use of Artificial Neural Networks (ANN) is being explored. In this article, the use of Fourier Descriptors for hand gesture recognition is discussed. The system model was set up as follows: Images of 24 hand gestures of American Sign Language (ASL) were subjected to edge detection algorithms, the contours were extracted and Complex Fourier descriptors were obtained as features for classification using a 2 layer feed-forward neural network The effects of subsampling of the contour, number of hidden neurons and training functions on the performance of the network were observed. Maximum accuracy of 87.5% was achieved.