Bag of Visualwords and Chi-Squared Kernel Support Vector Machine: A Way to Improve Hand Gesture Recognition

In this paper, we present an efficient and real-time technique using Bag Of Visual Words (BOVW) and chi-squared kernel Support Vector Machine (SVM) for recognizing static gestures of Arabic Sign Language alphabets. Thus, the images of static hand gestures are converted into DSIFT (Dense Scale Invariant Feature Transform) features and grouped using k-means clustering to create histograms. Then they are converted from non linear space into linear space using Chi-squared kernel. The result is fed into One-vs-All SVM classifier to build signs models. Training and test stages of this technique are implemented on hand postures images using cluttered backgrounds for different lighting conditions, scales and rotations. This technique of computing new hand features and modified classifier has a satisfactory recognition rate and can achieve good real-time performance regardless of the image resolution. As an application, it can be implemented on embedded system for real-time Human-Machine or Human-Human communication system.