Skip to content

Matlab App for Character Recognition from images of words using Neural Networks

Notifications You must be signed in to change notification settings

bhishmapitamah/Character-Recognition

Repository files navigation

Character Reognition

This is a matlab app for recognition of characters in a image containing a word. This project was done as a B.Tech Project with Ayush Mukund Gupta and Abhinav Rishikesh.

Preprocessing

-Abhinav Rishikesh

The input image is taken and is converted to binary form using otsu method.

Segmentation

-Ayush Mukund Gupta

The binarized image is segmented into individual characters using horizontal and vertical projections of the image.

Classification

-Ashutosh Pandey

Classification is done using Neural Network. The Neural Network is a single hidden layer network with 600 hidden layer nodes.

The input the Neural network is a matrix in which each row has 625 elements corresponding to the pixels of 25 x 25 image.

The output of the Neural network is matrix in which each row has 62 elements out of which only one element is 1 others are 0 denoting the chosen class [A-Z a-z 0-9].

The Code used for traning of the neural network is also provided in the Traning folder. The dataset used for training was a manually chosen subset of Erik's huge dataset.

https://erikbern.com/2016/01/21/analyzing-50k-fonts-using-deep-neural-networks.html

The neural networks hidden layer size, regularization parameter and optimizing function can be changed with ease but changing the number of hidden layers would take some understanding of code.

I coded the neural network while i was doing the coursera machine learning course by Andrew Ng. And some of the function are used from the course like 'fmincg.m' and 'displaydata.m'. The function fmin_adam is also from matlab file exchange.

About

Matlab App for Character Recognition from images of words using Neural Networks

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages