Rekognize

The problem

Recognizing a character means accounting for the many ways people draw the same shape. The project brings image preprocessing, trained recognition models, and a browser interface into one workflow.

What I built

Rekognize is a web application that identifies handwritten digits and alphabetic characters. It uses convolutional neural networks trained on MNIST and EMNIST, with TensorFlow handling prediction and Flask serving the application. The interface connects a drawing input to the recognition models.

• Use MNIST and EMNIST as the training data for digit and alphabet recognition.
• Build convolutional neural networks with TensorFlow and Keras.
• Preprocess drawings with OpenCV before passing them to the models.
• Connect the models to a Flask application and a drawing interface.

The result

A handwriting recognition experiment for digits and letters.