Marathi-Letter-Recognition-ConvNet

This project is Deep Learning Web Interfaced character recognition project. For sake of simplicity flask backend is used to minimize complexities. Basic working include you can draw the character on the canvas and system will detect the character.

Tools used :

• Jupyter Notebooks - Model Building , Data analysis , Exporting model
• Flask - Python Backend library for Simple Backend with python integrations
• Bootstrap - Clean Looking Frontend Web face
• DrawingBoard.js - Used to Build the Drawable Canvas element
• Javascript - Lightwieght JS used for DOM manipulation

Working :

• You draw the character on canvas
• Image Data URI is pulled to backed after you press predict
• Image Data URI is converted to PNG Image and downshaped
• Convolutional Model is loaded using loaded
• PNG image is shaped like model input i.e. 4D (1 ,? , ? , 1)
• Model is used to Predict the Output of Image passed
• Output and Image code (Generated by Server) is sent to Frontend in JSON
• Front End DOM updation using JS

Working Demo :