Character Recognition from Image Using TensorFlow

Abstract

Machine Learning (ML) is the study of pc algorithms that improve automatically through experience.  It is visible as a subset of  artificial intelligence. Machine gaining knowledge of algorithms build a  mathematical model  based on sample data, referred to as 'education statistics', in order to make predictions or decisions .Machine mastering algorithms are utilized in a wide style of programs encompass e-mail filtering, detection of network intruders or malicious insiders working towards a facts breach, handwritten individual recognition and pc vision.

Many practices have been made to classify information from handwritten statistics, however the laptop alone hasn’t been capable of classify statistics efficiently. Hence, the want of handwriting recognition system has grew to become up. In this undertaking we purpose at constructing a handwritten character popularity system using convolutional Neural Network. Which will be capable of understand the handwritten characters and provide the output using the training i.e (IAM) dataset.

Introduction

Understanding the handwritten characters or typed files is simple for people as we've the capacity to learn. The identical capability may be caused to the machines additionally with the help of Machine Learning . The field which deals with this problem is referred to as the OCR or additionally called Optical Character Recognition. It is the area of take a look at among diverse fields inclusive of recognizing of pattern. With the help of the OCR in banking area, legal scenarios, etc. Many vital and sensitive files may be processed quicker without human intervention. Hence, the want of handwriting popularity has come up. By schooling the laptop to apprehend the set of handwritten characters we will classify the handwritten characters and retrieve facts from them.

Introduction to Tensorflow

Tensor Flow is an open-source software library for dataflow programming which can perform various tasks. It is a symbolic math library, and also used for machine learning applications such as neural networks. It is used for both research and production at Google.

Tensor Flow was developed by the Google Brain team for internal Google use.

DATA ACQUISITION

The IAM Handwriting Database includes styles of handwritten English textual content which can be used to teach and take a look at handwritten text recognizers and to perform writer identification and verification experiments. The database changed into first published in[1] on the ICDAR 1999. The database carries styles of unconstrained handwritten text, which had been scanned at a resolution of 300dpi and stored as PNG photographs with 256 gray levels.

Operations

CNN:- The input image is fed into the CNN layers. These layers are educated to extract relevant features from the image. Each layer includes 3 operation. First, the convolution operation, which applies a filter out kernel of length 5×5 within the first layers and 3×3 inside the last 3 layers to the input. Then, the non-linear RELU feature is applied. Finally, a pooling layer summarizes picture regions and outputs a downsized version of the input.

RNN: - The feature sequence carries 256 features according to time-step, the RNN propagates relevant information thru this collection. The RNN output collection is mapped to a matrix of size 32×80.

CTC: - While training the NN, the CTC is given the RNN output matrix and the ground truth textual content and it computes the loss cost. Both the ground truth textual content and the diagnosed textual content can be at maximum 32 characters long.

Data(Input):- It is a grey-level photo of size 128×32. Usually, the pictures from the dataset do not have exactly this size,therefore we resize it (without distortion) till it both has a width of 128 or a height of 32. Then, we generate the photo into a (white) target picture of length 128×32.

Conclusion

The Convolutional Neural Network is used for the process of feature extraction and provide an appropriate output to the user. The neural community consists of several layers which assist in the education procedure .The error turned into reduced upto ‘12.065659%’ and accuracy advanced upto ‘71.252174%’.Hence,our Handwritten Text Recognition(HTR) system is generated the use of this NN model.

Future scope

• New features are often added to improve the accuracy of recognition.
• These algorithms can be tried on large database of handwritten text.
• The proposed work are further extended to work on degraded text or broken characters.
• Recognition of digits in the text, half characters and compound characters are often done to reinforce the word recognition rate.

References

1. Application of Neural Network In Handwriting Recognition by Shaohan Xu, Qi Wu, and Siyuan Zhang, Stanford University.
2. Official Website of Tensorflow. URL: www.tensorflow.org .
3. Wikipedia, the free online encyclopedia URL: www.wikipedia.org .
4. A Gentle Introduction to Adams Optimizer. URL: https://machinelearningmastery.com/adam-optimization-algorithm-for-deep-learning/

Did you like this example?

1162

61