Sign Language Recognition

The model developed can classify any letter in the American Sign Language (ASL) alphabet into the equivalent letter in the English Alphabet, which is a multiclass classification problem.

Using this model, we can easily convert the ASL into the respective English sentences.

You can view the deployed version of the model here

For a detailed explaination please refer this article

Model Characterstics

Input

A grayscale ASL letter image of shape (28X28)

Output

Equivalent letter in the English alphabet corresponding to the ASL input image

Note: The letters J and Z cannot be predicted as their corresponding representation in ASL requires motion.

Evaluation

Accuracy is used as the evaluation metric

Architecture

Workflow

1. Importing the necessary libraries
2. Preprocessing the input data
3. Defining the Model
4. Fitting the training data
5. Hyper Parameter Tuning
6. Selection of the best model
7. Testing the performance on the test set

Best Model After Hyperparameter Tuning

models/experiment-dropout-0

Tuned Hyperparmeters include

1. Convolution and Max Pooling Pairs
2. Filters in the convolution layers
3. Filter Size
4. Dropout

Hyperparameters that can be further considered

1. Batch Normalization - It normalizes the layer inputs
2. Deeper networks work well - Replacing the single convolution layer of filter size (5X5) with two successive consecutive convolution layers of filter size (3X3)
3. Number of units in the dense layer and number of dense layers
4. Replacing the MaxPooling Layer with a convolution layer having a stride > 1
5. Optimizers
6. Learning rate of the optimizer
7. Data Augmentation