|
| Ref./year | Dataset size and type | Classifier | Accuracy in % |
|
| [2]/2020 | 100 Arabic signs from regular ArSL | Euclidian distance | 95% |
|
| [4]/2016 | Static alphabet two datasets (dataset 1: 700 samples for 28 characters using bare hands and dataset 2: 700 samples for 28 characters using colored gloves) | C4.5 (J48), MLP, K-NN (IBK), and Naïve Bayesian classifiers | 80.67, 88.66, 90.7%, and 84.4% for dataset 1 and 89.5, 94.11, 97.5, and 96.63 for dataset 2 |
|
| [6]/2018 | 30 people shot actual images using smartphones. Volunteers gesture the 30 ArSL alphabets. Each letter uses 30 of the 900 images | SVM | 63.5% |
|
| [8]/2017 | The dataset has 25 words from unified ArSL dictionary | 3D CNN—softmax layer | 98% accuracy for observed data and 85% average accuracy for new data |
|
| [10]/2020 | ArSL2018 is comprised of a total of 54,049 images for the 32 ArSL alphabets and signs, which are gathered from 40 different participants | CNN | 97.6% |
|
| [12]/2021 | ArSL2018 contains 54,049 images of 32 sign language gestures | CNN | 97.29% |
|
| [13]/2020 | ArSL2018 contains 54,049 images distributed around 32 classes of Arabic signs | Deep CNN | 99% |
|
| [14]/2019 | ArSL2018 dataset consists of 54,049 images with 32 class | Transfer learning approach of deep CNN | 99.52% |
|
| [17]/2018 | 450 colored ArSL videos captured at a rate of 30 fps | Euclidean distance classifier | 97% |
|
| [18]/2011 | 6,000 sign images are obtained from six gestures | Max-pooling CNN (MPCNN) | 96% |
|
| [19]/2019 | 7869 images for recognizing 28 Arabic letters and numbers from 0 to 10 | CNN | 90.02% |
|
