Arabic Sign Language Datasets: Review and Improvements

The research on Arabic sign language recognition has increased dramatically in the last decade, which required the availability of high-quality datasets. This paper reviews and analyzes the available Arabic sign language datasets according to seven quality metrics where possible. Furthermore, this paper improves the best dataset according to the quality metrics by enhancing its labeling information and expanding the dataset using the augmentation method. The review results show that there are twelve datasets with only one available dataset for public research which is called ArASl2018. Further, ArASl2018 is improved in terms of attributes and expansion. The enhanced labeling gives the ability to apply the statistical analysis while expansion increases the accuracy as verified by the results.