Sign language recognition using Hilbert curve features

Abstract

In order to design an efficient vision-based gesture recognition system, the images of the hand gestures need to be represented in an accurate way. This paper presents a new method for representing hands' images based on the Hilbert space-filling curve. The method first segments the hand and then applies a Hilbert space-filling curve to extract a feature vector. Afterwards, a classifier, such as Support Vector Machine (SVM) and Random Forest (RF), are used to classify the gestures. The Hilbert curve representation is chosen in this work since it has proved its efficiency in representing shapes with uniform background, due to its localization-preserving property of pixels. Moreover, it is also known of being invariant to translation, scaling, and stretching. The results reveal the efficiency and suitability of the proposed approach compared to that of other works in real time. © 2013 Springer-Verlag.