Özet:

Abstract Sign language is used to communicate with deaf and dumb people; it is difficult for ordinary people to communicate with them. Hence, computer vision and automatic identification can reduce the difficulties of reaching them. Deep learning algorithms were used to distinguish sign language in different languages and styles. Convolutional Neural Networks (CNNs) are used in computer vision, particularly pre-trained algorithms. This research proposes using transfer and machine learning to distinguish Kurdish Sign Language (KSL). A KSL dataset was created to characterize the Kurdish language at the level of numbers and letters, using pre-trained algorithms for feature extraction and machine learning algorithms for classification. The proposed method was tested on two data sets; KSL and American Sign Language (ASL). The algorithms (VGG19 and RESNET101) are implemented in the feature extraction phase with pre-trained weights. The algorithms: Support Vector Machine (SVM), Decision Tree (DT), and Random Forest (RF), is dependent on the classification stage, and the CNN is designed for the KSL model. The efficiency of the proposed models is evaluated using (accuracy, recall, precision, and F1 score) metrics. The proposed model's outcomes illustrated that VGG19 is better than (RESNET101 and proposed CNN) algorithms in terms of feature extraction, and the random forest is the best classifier which achieved an accuracy rate of 95% at the numbers level and 97% at the level of the letter for KSL and ASL.

Anahtar Kelimeler: