Handwritten Tamil Character Recognition

Summary

Handwritten Tamil Character Recognition is a project to detect character written in Tamil a lingustic language script which is widely spoken in the state of Tamil Nadu. The applications of this OCR recognistion systems are preserving ancient tamil scripts, teaching how to write tamil etc.

The project has the Google inception-v3 as the model and is trained by the transfer learning approach. There is a website on which users can draw a character and it shows the top 5 characters that the model thinks it could be.

Goals

Build and train the Google inception-v3 classifer for recognising the characters

develop a website to collect data on written characters

develop a website with interactive UI for drawing the characters

deploy the system to cloud for training the model and hosting the site

Project Details

Software

Software packages used to build the project.

Software:

IDE: Python IDLE

Languages: HTML5, Bootstrap, JQuery, Python

Packages: scikit-learn, Tensorflow with tflearn, numpy, pandas

server: Flask

Prediction Model

The Google inception-v3 is a classifer used by google images for its search algorithm. Its a 13 Layer Deep convolutional Neural Network. and is available on tensorflow documentation pages for loading the weights.Its trained against ImageNet a big repo of various images. we will apply Transfer Learning to use this classifer to recognise Tamil Characters.

Results

Due to University norms I am not at the liberty to discuss the results of this project as this is still under working progress. what I can say is that the training approach was a success and we are still woking to improve the results of this classifer.