Upender-recognizer 0.0.3

Converting handwritten (digits) information to digital format

Real-time-handwritten-digits-recognition-using-Convolutional-Neural-Network

Description

Reading handwritten information like examination answer sheets is still a difficult task for many of us, because each one of us is having a different interpretation style. As the world is moving towards digitization, converting the hand written information to a readable digital format reduces the difficulty. This approach will be beneficial for the readers as it gives a better understanding of the information. With the help of machine learning and deep learning algorithms, the hand written patterns can be recognized and classify them accordingly to a digital format with human level accuracy.

This repository deals with predicting the real time handwritten digits only. To classify the handwritten digits MNIST data set is used for training the model. OpenCV python library is used for detecting the patterns in the real time handwritten digits. These detected patterns are predicted to human level accuracy with the help of a Convolutional Neural Network model.

Environment

• Colab
• Tensorflow
• Keras
• Numpy
• Pandas
• Matplotlib
• Sklearn
• Seaborn
• OpenCV

Data set

The MNIST data set (Modified National Institute of Standards and Technology database) has 70,000 handwritten images, which is divided into training data set of 60,000 images, and a testing data set of 10,000 images. Each digit in the data set is normalized and centered in a gray-level image with size 28 * 28, or with 784 pixels. All the pixels are stored in csv files, training files has 60,000 rows * 785 columns and testing files has 10,000 rows * 785 columns. Few sample images from MNIST data set are shown in the below figure.

Convolutional Neural Network Model

	Layers
CL1	Convolutional Layer 1
ML1	Max Pooling Layer 1
CL2	Convolutional Layer 2
ML2	Max Pooling Layer 2
CL3	Convolutional Layer 3
FL1	Dropout & Flatten Layer 1
DL2	Dense Layer 1
DL2	Dense Layer 2

Output

A real time handwritten scanned image was taken and uploaded in the Google Colab. The trained model recognized the digits from the image and displayed another image with digital numbers.