handwritten-digit

An handwritten recognition web app with the use of cgi-scripts

Start of the web app

Recognition of digit

This webapp consists of two major parts as shown below
1. UI logic
   1. index.html - the actual UI in HTML
   2. paint.js - Control of HTML Canvas that lets users to write the handwritten digit to be tested
   3. button-logic.js - handling the AJAX request to those Python CGI neural network scripts
   4. iteration-slider.js - responsible for the slider that shows training iteration
2. Python CGI script to handle the neural network for recognizing handwritten digit
   1. *-ubyte - actual training set of size 50,000 and test sample set of size 10,000 with total 60,000 28x28 images used by the neural network
   2. thetas.mat - the actual trained weight matrix used for prediction
   3. feedforward-prediction-cgi.py - loads the matrix in thetas.mat, and perform feedfoward prediction using the matrix.
   4. train-handwritten-digit-cgi.py - trains the weight matrix and save it to thetas.mat
   5. imshow-grayscale-mat.py - saves the raw pixel data under the CGI dir for debugging purpose.

Proof of concept
• Use of AJAX and JQuery to achieve rich interaction interface.
• Use of HTML Canvas and event-driven programming
• Use of Python CGI script
• Demonstration of practical use of neural network for basic image recognition

Notes
• One must configure their Apache server to enable the use of CGI script
• In order to have a working training functionality, timeout must be disable for the long loading time that might occur during the NN training stage