You are on page 1of 14

RECOGNITION OF TEXT IN TEXTUAL IMAGES

USING DEEP LEARNING

SAI HARSHITH.K-170701117
RISHIKESH.S-170701111
SANTHOSH.S-170701122
PROJECT SUPERVISOR: Mr. M.K VARADARAJAN
Asst. Professor, ECE dept,
SVCE.
OVERVIEW
• OBJECTIVE
• NEED FOR THE PROJECT
• LITERATURE SURVEY
• BLOCK DIAGRAM
• INTRODUCTION
• PROPOSED METHOD
• SOFTWARE USED
• PLAN OF WORK
• REFERENCES
OBJECTIVE
To recognize the text using deep neural network
 To classify whether a image is textual or non textual.
 To recognize the text if the image is textual.
NEED FOR THE
PROJECT
Information Retrieval

Information Filtering

Document Summarization
LITERATURE SURVEY
S.no Name of the Paper and Published Algorithm Used Efficiency Further
date Development
Distinguishing Text/Non-Text Combination of Convolutional
Natural Images with Multi- Neural Network and Multi- Recognition of text
1. 90% if the image is
Dimensional Recurrent Neural Dimensional Recurrent Neural
textual.
Networks (DEC-2016) Networks
TextField: Learning A Deep
Detection of text field using binary To improve the
2. Direction Field for Irregular Scene 80%
Text Detection (MAY-2019) classification efficiency.
Text and Non Text Scene Image Recognition of text
Classification for Visually Impaired
3. Binary Classification 99% if the image is
Through Alexnet Transfer Learning
Model (MAY-2019) textual.
BLOCK DIAGRAM
Convolutional
neural network Textual and Non
Input Image
(classifier) using textual images
Tntidb dataset

Image
Pre-Processing
Normalization Text recognition
(RGB image to
(changing pixel model
GRAY)
intensities)

Feature Extraction
Min Max Scalar
(Extracting pixel Recognized Text
(fitting of data)
intensities)
INTRODUCTION
What is Deep learning?
 Part of a broader family of machine learning methods based on artificial neural networks with
representation learning.
 Uses multiple layers to progressively extract higher-level features from the raw input.
DEEP NEURAL NETWORK

• An Artificial Neural Network (ANN) with


Multiple Layers between the Input and Output
Layers.
• Represents the type of machine learning when
the system uses many layers of nodes to derive
high-level functions from input information. 
TYPES OF CLASSIFIERS
Model Advantages Limitations
Rocchio Algorithm Easy to implement and computationally is The user can only retrieve a few relevant
very cheap. documents
Logistic Regression It does not require input features to be It cannot solve non-linear problems and
scaled (pre-processing) and also it does prediction requires that each data point be
not require any tuning. independent
Limited by data scarcity for which any
It works very well with text data and fast
Naive Bayes Classifier in comparing to other algorithms. possible value in feature space, a likelihood
value must be estimated by a frequentist.
Computational of this model is very
Effective for text datasets, non-parametric expensive and difficult to find optimal value
K-Nearest Neighbour and more local characteristics of text or
of k.
document are considered.
SVM can model non-linear decision Lack of transparency in results caused by a
Support Vector high number of dimensions (especially for
Machine boundaries and performs similarly to
logistic regression when linear separation. text data).
PROPOSED METHOD

Training &
Validation
Preprocessing

Dataset
SOFTWARE USED

Python

Database(TNTIDB)
JAN
DEC-JAN MID DEC Implementation of efficient
Study of Classifier Algorithms algorithm to classify textual and
non textual images

DEC-JAN MID
PLAN OF
WORK
FEB MAR
Study and Implementation of text Testing, Verification and
DEC-JAN MID from textual images
recognition Documentation
REFERENCES
• Pengyuan Lyu, Baoguang Shi, Chengquan Zhang, Xiang Bai , "Distinguishing Text/Non-Text Natural
Images with Multi-Dimensional Recurrent Neural Networks" in 2016 23rd International Conference on
Pattern Recognition (ICPR) Cancún Center, Cancún, México, December 4-8, 2016.
• Anilkumar B, Sreerama Murthy Velaga, A Aswani Devi, "Text and Non Text Scene Image Classification for
Visually Impaired Through Alexnet Transfer Learning Mode" in International Journal of Recent
Technology and Engineering (IJRTE) ISSN: 2277-3878, Volume-8, Issue-1, May 2019.
• Yongchao Xu, Yukang Wang, Wei Zhou, Yongpan Wang, Zhibo Yang, Xiang Bai, "TextField: Learning A
Deep Direction Field for Irregular Scene Text Detection" in International Journal on Document Analysis
and Recognition, May 2019
THANK YOU!

You might also like