Hand gesture recognition using discrete wavelet transform and support vector machine

In this paper a system to recognize static hand gesture is presented. The two dimensional wavelet transform is used for extracting features and the multiclass support vector machine is used for classification. The proposed system has 4 steps: 1) Image acquisition, 2) Image preprocessing, 3) Feature extraction and 4) Classification. The image is captured through digital camera, then converted to gray-scale, cropped and re-sized.

Two dimensional discrete wavelet transformation decomposition is applied on final image obtained after preprocessing, so that we get an approximate image of the 7th level as feature vector. This feature vector is an input to the SVM, which is first trained and then tested. The dataset is taken in real world scenario where several variations in terms of size, orientation, illumination are present within the same class of a gesture. This system has accuracy of 94% when trained and tested over 350 samples of 7 hand gestures. Also the system is able to tolerate salt and pepper noise up to 0.5 intensity without much compromising the accuracy.