MLP character recognition

Trains a multi-layer perceptron (MLP) neural network to perform optical character recognition (OCR).

The training set is automatically generated using a highly modified version of the captcha-generator node-captcha.

The network takes a 20x20-bit (400-bit) one-dimensional binary array as input and outputs an 8-bit binary array, which can then be converted into a character code. Initial performance measurements show promising success rates.

After training, the network is saved as a standalone module to ./network.js, which can then be used in your project with

var network = require('./network.js');
var output = network.activate(input);

Performance

abcdefghijklmnopqrstuvwxyz

Font: Arial, Helvetica, sans-serif
MLP specification:
- Neurons: (400 input, 40 hidden, 8 output)
- Learning rate: 0.1
- Training set:
  - Size: 52000 distorted characters
  - Sample:
Measured success rate: 98.52% (of 10000 random inputs from training set)

0123456789

Font: Arial, Helvetica, sans-serif
MLP specification:
- Neurons: (400 input, 40 hidden, 8 output)
- Learning rate: 0.1
- Training set:
  - Size: 20000 distorted characters
  - Sample:
Measured success rate: 99.79% (of 10000 random inputs from training set)

Usage

Clone this repository. The script is using canvas, so you'll need to install the Cairo rendering engine. On OSX, this can be done with the following one-liner (copied from canvas README):

$ wget https://raw.githubusercontent.com/LearnBoost/node-canvas/master/install -O - | sh

Then install npm dependencies and test it:

$ npm install
$ node main.js

Example

Here is an example run of the script:

$ node main.js
generating images ...
... done

neural network specs:
  layers:
    input: 400 neurons.
    hidden: 40 neurons.
    output: 8 neurons.
  learning rate: 0.1
  training set: 52000 distorted characters.

learning ...
progress: 10%
progress: 20%
progress: 30%
progress: 40%
progress: 50%
progress: 60%
progress: 70%
progress: 80%
progress: 90%
... done

network saved to ./network.js

testing on 10000 random input samples ...
progress: 10%
progress: 20%
progress: 30%
progress: 40%
progress: 50%
progress: 60%
progress: 70%
progress: 80%
progress: 90%
... done

success rate: 98.52 %

Todo

Add customizability (fonts, optional distortion, MLP specs, threshold etc.)
Add training based on target success rate
Add separate testing set to avoid possible bias

Contribute

Feel free to fork and submit pull requests.

wavelets / mlp-character-recognition Goto Github PK