Sign Language Alphabet Classification(CNN)

• 

  Correct prediction of Y

Inspiration

With this project, we wish to inspire more people to learn ASL and hopefully be interested in further exploration of this language.

Purpose

This algorithm classifies images into the corresponding English alphabet.

How we built it

We chose our dataset from Kaggle. Pandas, CSV, and Numpy libraries are used for loading and normalizing image data. Keras is used to implement, train, and test the CNN model. The performance of our model was determined by the accuracy score and classification report and the deployment of this CNN model was primarily done using Flask and pickle modules.

Challenges

Some challenges were finding the optimal hyperparameters during the training phase and building a web app from scratch with no previous experience.

Lessons & Achievements

We gained proficiency in image pre-processing procedures, implementation and evaluation details of CNN models, and navigation across multiple platforms. We are proud of the final accuracy score(96%), which out beats our initial goal for this project(95%).

What's next for Sign Language Alphabet Classification(CNN)

Future improvement includes the augmentation of training samples to obtain a more accurate prediction and we are also interested in the classification of ASL words and real-time ASL translation.

Built With

• flask
• kaggle
• python