LVQ FOR HAND GESTURE RECOGNITION BASED ON DCT AND PROJECTION FEATURES
Ahmed S. Tolba, Mohamed A. Abu Elsoud, Osama Abu Elnaser · 2009
The purpose of this paper is to set out an investigation into the use of handshape for the development of a gesture recognition algorithm. We present a system that performs automatic gesture recognition without using data gloves or colored gloves. Both the Discrete Cosine Transform and Projection features are used to match the input pattern against a standard gesture Database using a LVQ, algorithm. Experiments were based on 24 hand-shapes, produced by Thomas Moeslund [1]under special conditions and possible consideration of scaling, Translation, Rotation, Color and Illumination variance. A training set of 480 imaged (20 occurrence of each of the 24 hand-shape) and testing set of 480 images (20 occurrence of each of the 24 hand-shape) has been used. The correct recognition rate is approximately 83.13 % to 84.17 % for diagonal projection features and approximately 85.36 % to 87.50 % for Cols & Rows projection features. It reaches 91.04 % for DCT features. K e y w o r d s: gesture recognition, discrete cosine transform