TECHNIQUES OF TEXT

DETECTION AND EXTRACTION

An Overview of Different Techniques and thier
Implementations

VIVEK RANA
1709113908
 INTRODUCTION
● The text detection and extraction are the techniques of patterns
detection..

● Text remains a challenging trouble due to the variety of shades,

fonts, colours as well as the presence of complex backgrounds.

● Text detection and recognition in images which aims at integrating

advanced OCR and text-based searching technologies.

● Text Recognition finds its application across various domains like AI,
Multimedia, Robotics and much more.
 INTRODUCTION
➔ Basic model of any Text Detection

and Recognition system.

➔ Consists of Text Localization,

Verification, hypothesis generation

and Result Selection.

➔ First, It localizes the Text Line

,Extract the features and Binarization

takes place followed by CCA.

Fig 1: Basic Model of Text Detection & Recognition

Picture Source: [2]

 LITERATURE SURVEY
Author(s) Title Year Techniques Used Conclusion

If Binarization
result is poor,
Text Recognition
the chance of
K.Wang, B. using Part based
PART BASED TREE correctly
Babenko and Tree structured 2011
STRUCTURE recognizing the
S.Belongie character
text is quite
Detection .
small.

Achieves
Cong Yao, Xiang Text Detection of significant
TWO-LEVEL
Bai, Wenyu Liu , Arbitrary performance on
2012 CLASSIFICATION
Yi Ma, Zhuowen Orientations in texts of arbitrary
SCHEME
Tu Natural Images orientations with
50% Accuracy.
Table 1.1: Literature Survey(1)
 LITERATURE SURVEY
Author(s) Title Year Techniques Used Conclusion

State-of-the-art
End-to-End Text
Tao Wang, David MULTILAYER NEURAL performance on
Recognition with
J. Wu, Adam NETWORKS WITH standard
convolutional 2014
Coates, Andrew UNSUPERVISED benchmarks
Neural
Y. Ng FEATURE LEARNING with 60%
Networks(CNN)
Accuracy.

Table 1.2: Literature Survey(2)

 LITERATURE SURVEY
Author(s) Title Year Techniques Used Conclusion

Text-Attentional
Improving the
Tong He, Weilin Convolutional
SPATIAL IMAGE state-of-the-art
Huang, Yu Qiao, Neural Network 2016
FILTERING results with a F-
and Jian Yao by training Text-
measure of 0.82
CNN

Achieved high
A robust recall, and a fast
L. Sun, Q. Huo, approach for subsequent
TEXT BASED IMAGE
W. Jia, and K. text detection 2017 filtering stage
RETRIEVAL
Chen from natural for
scene images improving
precision.

Table 1.3: Literature Survey(3)

 TECHNIQUES OF TEXT DETECTION
● SVM(Support Vector
Machine) Algorithm,

● Image Filtering Algorithms

 Spatial Filter

 Temporal Filter,

● CNN (Convolutional Neural

Network) Algorithm,
 CONTINUED...
● SVM(Support Vector Machine) : SVM is a supervised machine
learning algorithm which can be used for classification or regression
problems. It uses a technique called the kernel trick to transform
your data .
● Image Filtering : A filter can be applied to reduce the amount of
unwanted noise in a particular image and can be used to reverse
the effects of blurring on a particular picture.
A.) Spatial Filtering, B.) Temporal Filtering.
● CNN(Convolutional Neural Networks) : The main purpose of the
convolution step is to extract features from the input image.
 TECHNIQUES OF TEXT EXTRACTION

● Compression Based Algorithm

● Corner Response Based Method,

● Edge Detection Algorithm,

● Nearest Neighbour Clustering Based Method (NNC)

 METHODS USED FOR TEXT DETECTION
AND RECOGNITION

Table 2: Different Methods and their comparison

Table Source: [6]

APPROACHES OF TEXT DETECTION AND
RECOGNITION
APPROACHES OF TEXT DETECTION AND
RECOGNITION
 APPLICATIONS

★ Banking (To read Credit ★ Government Sector (Form

Cards). Processing).

★ Libraries (To Convert ★ Used in Car number plate

Scanned images into Text). Recognition System.

★ Used in Recognition of various

★ Undesirable Text Removal
Languages’ .
From images
 CONCLUSIONS
● Accuracy depends on performance of Preprocessing and
Binarization.

● AI approach is suitable for Text Detection and Recognition systems.

● CNN helps in extracting useful features from Image.

● Connected Component Analysis helps in aligning the text.

● ANN help in achieving higher accuracy and recognition of various

Types of Text.
 REFERENCES
● [1] K. Wang, B. Babenko, and S. Belongie, “End-to-end scene text
recognition” 2011, in IEEE International Conference on Computer
Vision (ICCV).
● [2] C.Yao, X.Bai, W.Liu, Yi Ma, Z.Tu ― “Detecting Texts of Arbitrary
Orientations in Natural Images’’,2012,IEEE Part IV, LNCS 8692, pp.
497–511
● [3] T.He, W.Huang, Yu Qiao, and Jian Yao, “Text-Attentional
Convolutional Neural Network for Scene Text Detection”, 2016, IEEE
Transactions on Image Processing.
● [4] T.Wang, David J. Wu, A.Coates, Andrew Y. Ng, “End-to-End Text
Recognition with Convolutional Neural Networks”, 2013, ICPR
 REFERENCES
● [7] M.Jaderberg, K.Simonyan, A.Vedaldi, A.Zisserman,‖ Reading Text
in theWild with Convolutional Neural Networks‖, 2016, IEEE, Int J
Comput Vis (2016),
● [8] W.Huang, Y.Qiao, and X.Tang, “Robust Scene Text Detection with
Convolution Neural Network Induced MSER Trees” , 2014, IEEE Part
IV, LNCS 8692, pp. 497–511
● [9] X.Cheng Yin, W.Yi Pei, J.Zhang and H.Wei Hao, “Multi-Orientation
Scene Text Detection with Adaptive Clusteri.ng” , 2015, IEEE
Transcations on Pattern Analysis and Machine Intelligence.
 REFERENCES

● [10] Z.Zhang, W.Shen, C.Yao, X.Bai, “Symmetry-Based Text Line

Detection in Natural Scenes”, 2015, IEEE Computer Vision and
Pattern Recognition (CVPR).
● [11] B.Singh, and R.Maini, “Skew Detection and Correction of
Gurmukhi Words from Natural Scene Images”, International Journal
of Signal Processing, Image Processing and Pattern Recognition.,
Vol.9, 2016.
● [12] C. Yao, X. Bai, and W. Liu, ― “A unified framework for multi
oriented text detection and recognition” Image Processing, IEEE
Transactions on, vol. 23, no. 11, pp. 4737 4749, 2014.
 REFERENCES

● [13] K. Simonyan and A. Zisserman, ― “Very deep convolutional

networks for large-scale image recognition”, In 2015, International
Conference on Learning Representation(ICLR).
● [14] Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S.
Guadarrama, and T. Darrell, ― “Caffe: Convolutional architecture for
fast feature embedding’’ In 2014, in Proceedings of the ACM
International Conference.
● [15] L. Sun, Q. Huo, W. Jia, and K. Chen, ― “A robust approach for text
detection from natural scene images” ,Pattern Recognition,vol.48,
pp.2906–2920,2017.