Shape Based Pakistan Sign Language Categorization Using Statistical Features and Support Vector Machines

Abstract

All over the world, deaf use sign language to communicate each other. The signs are built and communicated through movements and the shapes of the hands. Pakistan sign language (PSL) is used by the deaf community of Pakistan. Automatic recognition of PSL alphabets into the predefined categories is presented here. The seven categories are defined using the visibility, shape, and the orientations of both the fingers and the hand. The histograms of the uniform local binary (lbp) for the neighboring distance of one, two, and three are computed and concatenated to form a single vector. At a later step, six statistical features of the combined histogram are computed, i.e., standard deviation, variance, skewness, kurtosis, entropy, and energy. The classification is achieved using support vector machines (SVMs). The modality of one-verses-one is used for the adoption of binary SVM into the multi-class SVM. The proposed technique is validated over the data set of 3414 PSL signs, taken through the help of seven native signers. The performance of the proposed technique is evaluated by precision, recall, accuracy, and f-measure, while it is elaborated through the classification matrix, tabular, and graph representations.