Sunteți pe pagina 1din 11

HANDWRITTEN

DIGIT REGONIZER
USING NEURAL NETWORKS AND SVM

MENTOR
Mr. Harsh Taneja

STUDENTS
Shubham Suyal
Kunal Gusain

OBJECTIVE
To successfully implement neural networks and
SVM to make a digit recognizer with high
accuracy. The idea is to take a large number of
handwritten digits, known as training examples,
and then develop a system which can learn from
those training examples. In other words, the
neural network uses the examples to
automatically infer rules for recognizing
handwritten digits.

INTRODUCTION
Digit recognition system is used for recognizing
the digit in various font types, and/or for
handwritten digits and converting them into
digital numerals. This project is a combination of
digit detection and recognition. This application
uses the input image to detect digits given by
the user and perform basic operations
accordingly

Modules
Module

1: Loading of MNIST training data.

We used the MNIST data set, which contains tens of thousands of


scanned images of handwritten digits, together with their correct
classifications.
Module

2: Representation of neural network.

The wordnetworkin the term 'artificial neural network' refers to


the interconnections between the neurons in the different layers
of each system. A system has three layers. The first layer has
input neurons which send data via synapses to the second layer
of neurons, and then via more synapses to the third layer of
output neurons.

Modules
Module

3: Applying SGD.

Stochastic Gradient descentis afirstorderoptimizationalgorithm. To find alocal minimumof a


function using gradient descent, one takes steps proportional to
thenegativeof thegradient (or of the approximate gradient) of
the function at the current point.
Module

4: Implementing SVM and making GUI.

Using Support Vector Machine, SVM as a classifier makes


performance better than neural networks in terms of accuracy.
Also, we take the input from the user, that is, the digit drawn by
the user which our recognizer predicts the value of (GUI).

PROGRESS LAST TIME

We had been successful in making a simple digit recognizer


using predefined data (digits) for recognition.

The dataset we used for our project was MNIST dataset.

We had implemented a simple neural network and trained


it from the given dataset to predict the given digit.

We also had implemented Stochastic Gradient Descent


Algorithm instead of regular logistic regression algorithm to
train the neural network.

DEMO SCREENSHOTS

PROGRESS THIS TIME

We used applied SVM as the accuracy check of neural


network.

Now we have made a simple GUI for the digit recognizer.

The GUI now

Loads the mnist data into system,

Trains the machine from mnist dataset,

Takes the digit from the user, and,

Displays the predicted digit.

DEMO SCREENSHOTS

CONCLUSION

The digit recognizer will recognize the handwritten


digit in the image which will be provided from the
user.

The future scope includes using Support Vector


Machines(SVM) instead of neural network to improve
the accuracy of the program.

Future scope also includes making a GUI for user


friendly interface.

THANK
YOU