Chars74k_Recognition

Character Recognition of Chars74k dataset - Handwritten and Images

using Histogram of Oriented Gradients and Neural Networks

Please visit HERE for analysis and results.

The code is divided into two parts-

1. C++ - Preprocessing input images and feature vector formation

2. Python/MATLAB - Using the features to train Neural Network

Preprocessing and feature vector formation code is "proj.cpp" in "Code" folder.

"Neural Network" folder contains two codes : in MATLAB and in Python (Scikit-learn)

"nn-hand.py" and "nn-img.py" are python codes.

"imgANN.m" and "handANN.m" are MATLAB codes.

NOTE: Python code requires 'Scikit-learn' library. It is preferable to install 'Anaconda' distribution of python to run it.

The data (images) used are not provided since they are very large in size. Download the feature vector .txt files and original images used here:

https://drive.google.com/drive/folders/0B6TI76M3_KiyZUR0UE1WRXJka1E?usp=sharing

Preprocessing code:

Compile:

$ cmake .
$ make

An executable name 'proj' is generated.

Run using:

$ ./proj

NOTE: To run the code 'data-img' and 'data' directories must be present alongside the 'code' directory. 'data-img' contains the 'Image Characters' 'data' contains 'Handwritten Characters'

These folders are very large. Please download them from the shared folder and place them alongside 'code' folder.

The data used is taken from the 'chars74k' dataset available at: http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k/

This dataset is converted to .png images for each class (26 lowercase, 26 uppercase and 10 digits)

Labels 1-10 = Digits Labels 11-36= Uppercase Labels 37-62= Lowercase

We use these images, read them, preprocess them, then form a feature vector which is stored in file "features.txt".

This "features.txt" is given as input to Neural Network implemented in Python using Scikit-learn library. The neural network trains on some percentage of images and tests on the rest.

After features.txt is generated place it along with neural network code.

"nn-hand.py" and "nn-img.py" are python codes.

Run using:

$ python nn-hand.py -- for handwritten characters

The code reads data from 'features-handHOG.txt' and 'features-handPIX.txt'. Please place them alongwith .py file before running. Accuracy is displayed after model is trained.

Run using:

$ python nn-img.py  -- for image characters

The code reads data from 'features-imgHOG.txt' and 'features-imgPIX.txt' Please place them alongwith .py file before running.

Please download feature-______.txt files from shared link:

https://drive.google.com/drive/folders/0B6TI76M3_KiyZUR0UE1WRXJka1E?usp=sharing

---

"imgANN.m" for image characters and "handANN.m" for handwritten characters are MATLAB codes.

Run using MATLAB

Please feature-____.txt files alongwith .m file before running.