Digit Recognition Using Machine Learning and Convolutional Neural Network

The purpose of the sign language recognition system is to establish communication with people having hearing and speech problem. The proposed system is implemented for recognition of 0-9 digits. In this paper, we compare the performance of two methods i.e. contour-SVM based method and convolutional Neural Network (CNN) method under different conditions like rotation, scaling with a constant background. These methods are applied on three standard databases-SLD, ASL and ASL-FS. In the contour-based method, we find the contour and draw convex hull. Gestures are recognized on the basis of length and angle of convex hull and SVM is used for classification purpose. CNN based method uses five convolutional layers are used to classify the digit data. The proposed algorithm achieves accuracy 69% for contour-SVM and 98.31% for CNN based approach.