Scribd
Upload
29 views19 pages

Handwritten Digit Recognition

Uploaded by

prathamxyz783
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
29 views19 pages

Handwritten Digit Recognition

Uploaded by

prathamxyz783
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 19

HANDWRITTEN DIGIT

RECOGNITION
CREATING A SYSTEM THAT RECOGNIZES HANDWRITTEN DIGITS (0-9) USING
MACHINE LEARNING ALGORITHMS AND CNN MODEL WITH THE HELP OF
PYTHON PROGRAMING LANGUAGE AND ITS LIBRARIES.

PRATHAM RAJBHAR - 20105A0022


ROHIT MASMAR - 21105A2001
VINEET PATIL. - 20105A0045
AIM AND OBJECTIVE

Aim: To develop a robust system capable of accurately recognizing handwritten digits (0-9) using
machine learning algorithms.
Objective:
1. Gather a diverse dataset of handwritten digits.
2. Preprocess and standardize the dataset for model training.
3. Create a model with machine learning algorithms like SVM, CNN, and Random Forests.
4. Evaluate models based on accuracy, precision, recall, and F1 score.
5. Optimize model performance through fine-tuning and validation.
6. Develop a user-friendly interface for digit input and output.
SOFTWARE REQUIREMENTS

VS code Python
PYTHON LIBRARIES USED

• Pygame
• Numpy
• Tenserflow
• Keras
• Matplotlib
• Open CV
MNIST DATA SETS AND .TFF FILES
MNIST: The MNIST dataset is a widely used benchmark dataset in the
field of machine learning and computer vision. It consists of 70,000
grayscale images of handwritten digits (0 through 9), each of which is a
28x28 pixel image. The dataset is split into a training set of 60,000 images
and a test set of 10,000 images, making it suitable for training and
evaluating machine learning algorithms for digit recognition tasks. MNIST
is commonly used for tasks such as classification, where the goal is to
correctly identify the digit represented in each image. It has been
instrumental in benchmarking the performance of various machine
learning models and algorithms in the research community.

TFF File: FreeSansBold.tff file is use for UI Graphic fonts to write


dgits on display surface of system built
Mnist data sets
PROPOSED SYSTEM
STEPS TO BUILD MODEL

Step 1: Acquisition and preprocessing of MNIST data


Step 2: Create and compile CNN model
Step 3: Train the model with processed data
Step 4: Evaluate the trained model accuracy with test data
Step 5: Save the train model in .h5 file
WHAT IS H5 DATA FORMAT?

An H5 is one of the Hierarchical Data Formats (HDF) used to


store large amount of data. It is used to store large amount
of data in the form of multidimensional arrays. The format is
primarily used to store scientific data that is well-organized
for quick retrieval and analysis.
BUILDING THE SYSTEM

Step 1: Create UI to have board for writing digit


Step 2: Create array variables to get handwritten digit image data features
Step 3: Resize, preprocess and normalise the input data to match MNIST data
Step 4: Load CNN Model for prediction and analysis of the input data
Step 5: Run the model on the processed data set for prediction
BASIC CNN ARCHITECTURE
A Convolutional Neural Network (CNN) is a type of deep neural network commonly used for image
recognition tasks like handwritten digit recognition. Here's a basic explanation of CNN architecture for
this task:

 Input Layer: The input layer of the CNN takes the handwritten digit image as input. Typically, these
images are grayscale and have a fixed size, such as 28x28 pixels for MNIST dataset.
 Convolutional Layers: Convolutional layers are the core building blocks of CNNs. They consist of a
set of filters (also called kernels) that slide over the input image, performing element-wise
multiplication with the local region of the input and then summing the results to produce a feature
map. Each filter captures different features of the input image, like edges, textures, or patterns.
Convolutional layers are responsible for extracting relevant features from the input image. As the
layers progress, they typically learn more complex and abstract features.
 Activation Function: After each convolution operation, an activation function like ReLU (Rectified
Linear Unit) is applied element-wise to introduce non-linearity into the network. This allows the
network to learn complex patterns and relationships in the data.
 Pooling Layers: Pooling layers are used to downsample the feature maps generated by the convolutional
layers, reducing their spatial dimensions (width and height). Max pooling is a common technique where the
maximum value within each local region of the feature map is retained, discarding the rest. Pooling helps in
reducing computational complexity, controlling overfitting, and making the learned features more
translationally invariant.

 Fully Connected Layers (Dense Layers): After several convolutional and pooling layers, the remaining
feature maps are flattened into a one-dimensional vector and fed into one or more fully connected (dense)
layers. These layers perform high-level reasoning on the extracted features and map them to the output
classes, which in the case of handwritten digit recognition are the digits 0 through 9.

 Output Layer: The output layer typically consists of a softmax activation function, which converts the raw
output of the previous layer into probabilities for each possible class (0-9). The class with the highest
probability is chosen as the predicted digit.

During the training process, the network learns to adjust the weights of the filters and neurons through
backpropagation and gradient descent, minimizing a loss function such as categorical cross-entropy, which
measures the difference between the predicted probabilities and the actual labels. This process continues
iteratively until the network achieves satisfactory performance on a validation set.
MODEL BUILDING AND TRAINING CODE
Model Application and prediction code
RESULT:

MODEL TEST ACCURACY: 80%

ACUAL ACCURACY: 20-40%


CONCLUSION
It is observed that actual accuracy of the system and test accuracy of
model are way different and there is also inconsistency in the prediction
result, this is occurring due to following reasons:
• Dimensions and contrast of Train and test data on which model is trained and tested
are different from those of input data provided by user
• Reshaping such input data may have normalized it to similar to MNIST Data set but
may loose or alter the information of input data in the process leading to errors in
prediction
• There may be uneven population of similar data such as having no. of MNIST DATA
of digit like 8,3,4 more while having data of digit like 6,1,2 least leading to biases in
the prediction and inconsistency

You might also like