Alphabets Arabic Sign Language Recognition Based on A Hybrid Model Combining Linear Discrimination Analysis and A One-Dimensional Convolutional Neural Network

Abstract

     Due to the increasing number of people suffering from hearing impairment in Arab countries, automatic sign language translation has become a pressing necessity to reduce the gap between hearing impairment people and the community, hence minimizing their isolation. In this paper, we provide a new proposal for an Arabic Sign Language (ArSL) detection and recognition system capable of localizing and recognizing ArSL alphabets via a merge of features extracted using a Linear Discriminant Analysis (LDA) algorithm and a one-dimension Convolutional Neural Network (CNN), which is a new method to our knowledge. The important parameters used in this proposal are measured to select the best. The accuracy of this proposal was about 99.98%. Also, this model worked well with some of the challenges related to the detection of sign languages, such as variation of image illumination and background. Finally, comparing the results with other works prove the robustness of this proposal