SlideShare a Scribd company logo

Texture features based text extraction from images using DWT and K-means clustering

Download as PPTX, PDF
D
Divya Gera

This document presents a multilingual text extraction and recognition system based on texture features and a k-means clustering algorithm, addressing challenges in varying text size, style, and complex backgrounds. It outlines the methodology, experimental results, and the system's high performance metrics, including a detection rate of 98.96%. The proposed technique shows potential for further improvements and applications in aiding visually impaired individuals and translating text across languages.

Data & Analytics
Related topics:
k-means-clustering
1 of 64
Downloaded 13 times
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
TEXTURE FEATURES BASED TEXT EXTRACTION AND RECOGNITION SYSTEM Guided By:- Dr. Neelu Jain Associate Professor PEC University of Technology Presented By:- Divya Gera 13207003 ME(Electronics)
Contents  Introduction  Motivation  Objectives  Literature Survey  Proposed Methodology  Experimental Results  Conclusion  Future Scope  References
Introduction
 Text content contains high level of semantic information as compared to visual information.  It provides important contents for information indexing and retrieval, automatic annotation and structuring of images.  Text extraction involves detection, localization, binarization, extraction, enhancement and recognition of the text from the given image.  These text characters are difficult to be detected and recognized due to their deviation of size, font, style, orientation, alignment, contrast, complex colored, textured background
Types of text images i.) Document Images
ii.) Scene Text Images
iii.) Caption Text Images
Process of Text Extraction & Recognition
Applications of text extraction and recognition  Document indexing and retrieving Vehicle license plate detection
Identification of parts in industrial automation Postal code from address on the envelope
Street signs Libraries for computerized storage of books
Bank cheque processing Blind and visually impaired person
Text extraction techniques Classification of text extraction techniques
 Region-based method uses the properties of the color or gray scale in the text region or their differences to the corresponding properties of the background.  Edge based method is focused on high contrast between the text and the background.  CC-based methods use a bottom-up approach by grouping small components into successively larger components until all regions are identified in the image.  Texture based method uses the fact that text in images has discrete textural properties that distinguish them from the background.
Motivation  Text in multimedia documents or images contains important information for visual content understanding and information retrieval.  It is very challenging to design a general-purpose text extraction system due to the variations in text due to font size, style, color, orientation, and alignment.  In case of scene text images problem of low contrast, blurring, reflection, uneven illumination shadow, and perspective distortion exists along with text variations.  Also, less work has been done on multilingual text extraction.
Objectives  To design and develop an algorithm for multilingual text extraction based on DWT coefficients, texture features, and k- means clustering algorithm.  To overcome the problems like uneven illumination and shadow, blurring and scratching, multi color, and complex background in the scene text extraction.  Develop a text recognition system for English language.  Compare the performance of the system in terms of parameters like Detection Rate, Precision Rate, and Recall Rate with the existing systems.  Developing a GUI for proposed algorithm.
Literature Survey
Author (year) Technique used Images Parameter Remarks Yao et al. (2007) CC and Support Vector Machine (SVM) Complex background images PR=64% RR=60% Pixels of each character assumed to have similar color. Lai et al. (2008) Edge detection and K- means clustering Signboard Images Efficient for uneven illumination Song et al. (2008) Histogram Projection and color based K- means clustering Chinese text PR=77.05% RR=75.63% K=3 gives best performance
Author (year) Technique used Images Parameter Remarks Dinh et al. (2008) Edge detection and Histogram Projection Signboard Texts Low complexity algorithm. Fan et al. (2009) Stroke features and connected component Caption text images PR=95.2% RR= 94.5% Color information is not fully used Audithan et al. (2009) Haar DWT, logical AND operator, Dynamic thresholding Document images DR =94.8 % Independent of contrast. Angadi et al. (2010) Discrete Cosine Transform and texture features extraction Natural scene images DR=96.6% Inefficient for complex background
Author (year) Technique used Images Parameter Remarks Anoual et al. (2010) Edge detection, texture features, connected compo- nent analysis Complex back-ground images PR=95% RR=89% Robust and effective. Kumar et al. (2010) CC Analysis ICDAR scene images PR=90% RR=89% Multilingual Text extraction. Hassanz- adeh et al. (2011) Morphological operator, Decision classifier Logo in document images PR=95.6% Accuracy=86. 9% A novel and fast method for logo detection. Chandra- sekaran et al. (2012) Morphological operation, and SVM ICDAR 2003 dataset PR=95% RR=92% Fails in case of high illumination
Author (year) Technique used Images Parameter Remarks Zaravi et al. (2011) DWT, Dynamic thresholding, Region of Interest (ROI) Colored books and journal covers DR=91.20% Robust to noise. Zhang et al. (2012) Edge Enhancement and CC Web and caption text images DR=92.4% Insensitive to various types of background noises. Seeri et al. (2012) Median filter, Sobel edge detector, connected component labeling, order static filter. Kannada text images PR=84.21% RR=83.16% Accuracy = 75.77% Fails to extract very small characters.
Author (Year) Technique used Images Parameter Remarks Azadboni et al. (2012) FFT Domain Filtering , SVM Classification, K-means clustering Scene text images DR= 98.10% Characters having uniform colour. Anupama et al. (2013) Morphology operators, Histogram Projection ( X and Y histogram) Handwritten Telugu document images. DR=98.54% Accuracy =98.29% Fails in case of touching characters & overlapping lines. Raj et al. (2014) CC based Natural Scene Images (Devanagari) PR= 72.8% RR=74.2 % Fails for small slanted/ curved text.
Proposed Methodology
Block diagram of text extraction
Pre-processing  Although the color component may differ in a text region, yet it does not provide any information for text extraction.  Also, the processing of three components in the RGB image is difficult.  Colored input image is converted into gray scale image. Input image Gray Scale image
2D-DWT  The level-1 2D DWT wavelet has been applied decomposition to gray-scale images. It decomposes the image into four sub- bands: 1 approximation sub band and 3 detailed sub bands.  LL sub-band: Horizontal and vertical directions both are at low frequencies.  LH sub-band: Horizontal direction is at low frequency and vertical direction is at high frequency.  HL sub-band: Horizontal direction is at high frequency and vertical direction is at low frequency.  HH sub-band: Horizontal and vertical directions both are at high frequencies.
2D- DWT sub-bands (a) Approximation image (b) Vertical edges (c)Horizontal edges (d) Diagonal edges
Feature extraction through sliding window  A small overlapped sliding window ( m×n) is scanned over each high frequency sub bands.  Zero padding is done, if required.  The text area has irregular texture property to a certain extent, so the text area can be looked as the special texture.  Language independent statistical features i.e. mean and standard deviation are calculated for the high frequency sub- bands.
K-means clustering algorithm  Unsupervised technique of classification.  Divides the set of points into k clusters so that the intracluster similarity is high but intercluster similarity is low.  This is done by minimizing the sum of distances (euclidean distance) between the points and the cluster centers.  The clustering of image is done on the basis of texture features of LH, HL, and HH sub-bands.  This algorithm divides the image into k=3 clusters i.e. simple background, complex background and text clusters. The cluster that has the higher mean and standard deviation values is the text cluster.  For simple background, image is divided into 2 clusters i.e. background and text cluster.
Morphological filter  The text cluster is mapped on a mask image by replacing the pixels in the text cluster by 1s and background cluster by 0s.  Morphological dilation operation is employed to fill the gap in the text region.  Dilation, basically, adds pixels to the boundaries of objects in an image. The number of pixels added to the objects in an image depends on the size and shape of the structuring element used to process the image.  In case of complex background, there may be some non text region in the mask image which is needed to be filter out.
Steps involved in text extraction for simple background image (a) Input image, (b) Gray scale image, (c) Output of 2D DWT, (d) Background cluster, (e) text cluster, (f) Extracted text image. (c)(a) (b) (d) (e) (f)
Steps involved in text extraction for complex background image (a) Input image (b) Gray scale image (c) Output of 2D DWT (d) Simple background cluster
(e) Complex background cluster (f) Text cluster (g) Extracted text output
Block diagram of text recognition
 A template file has been created in which the letters A-Z, a-z, and numbers 0-9 are stored in the form of images.  A horizontal projection profile technique is used to isolate each line of the text.  The lines are segmented in the words by again scanning the image horizontally and detecting the spaces in between the words.  Each line is scanned vertically to detect and isolate each character within the line.  Then the correlation of each character in the target image with the template file character is found.
Experimental Results
GUI of the proposed technique (a) Input image
(b) Pre-processing
(c) Mask image
(d) Morphological filter
(e) Extracted text image
Text extracted from caption text images (a) multicolor text (b) text of different size
Text extracted from outdoor scene images
Extracted text output from indoor scene images
Text extracted from scene images with complex background
Text extracted from image with curved text Text extracted from image with tilted text
Text extraction in Punjabi (a) Multicolor document image (b) Caption text image
Text extracted from multilingual images
Parameters for performance evaluation of text extraction  Detection Rate (DR)= Ratio of the text regions correctly detected by the algorithm to the total number of text regions.  Precision Rate (PR)= Ratio of correctly detected characters to the sum of correctly detected characters plus false positives.  Recall Rate (RR)= Ratio of the correctly detected characters to sum of correctly detected characters plus false negatives.
Dataset DR PR RR ICDAR and Kaist 98.93% 97.69% 97.83% Own dataset 98.99% 97.91% 98.27% Wavelets DR PR RR Haar 0.9979 0.9850 0.9835 db2 0.9942 0.9653 0.9497 bior1.3 0.9891 0.9618 0.8876 sym3 0.9894 0.9595 0.9131 coif1 0.9873 0.9364 0.9070 Comparative analysis of different wavelets Performance analysis of proposed text extraction technique
Authors Method DR PR RR Angadi et al., 2010 Discrete Cosine Transform and texture features extraction 96.6% - - Kumar et al., 2010 Connected Component Analysis - 90% 89% Azadboni et al., 2012 FFT Domain Filtering , SVM Classification, K-means clustering 98.10% - - Proposed Method DWT and K-means Clustering 98.96% 97.8% 98.05% Comparison of proposed text extraction technique with others
Output of text extraction and recognition system Sr. No. Input Image Extracted text output Recognized Text 1. 2.
Sr. No. Input Image Extracted text output Recognized Text 3. 4.
Parameter for text recognition performance evaluation Total Characters Falsely detected characters Accuracy (%) Proposed method 1000 49 95.10 Accuracy of text recognition system is defined as the ratio of correctly recognized characters to the total number of characters Accuracy of proposed text recognition system
Conclusion  Text is successfully extracted from complex color images i.e. document text images, scene text images, caption text images, multi-colored text images, newspaper images, book cover images etc.  The proposed technique is insensitive to font size, style, color and alignment etc.  Independent of language.  The proposed method show that this method gives promising results in terms of DR (98.96%), PR (97.8%), RR (98.05%), and accuracy (95.1%) as compared to other methods.
Future scope  Even though the performance of proposed system is excellent, there is always scope for improvement.  The proposed text extraction technique is inefficient for very low contrast and high illumination images.  The proposed system along with text to speech converter will be helpful to blind people.  The text after extraction and recognition can be translated from one language to other and will help the people to translate the text written on sign boards, street names etc. to their native language.
References  Angadi, S.A., & Kodabagi, M., Text region extraction from low resolution natural scene images using texture features, IEEE 2nd International Advance Computing Conference, Patiala, India, Feb 19-20, 2010, pp. 121-128.  Anoual, H., Aboutajdine, D., Ensias, S.E., & Enset, A.J., Features extraction for text detection and localization, 5th International Symposium on I/V Communication and Mobile Network, Rabat, Sept. 30- Oct.2, 2010, pp. 1-4.  Anupama, N., Rupa, C., & Reddy, E.S., 2013, Character segmentation for telugu image document using multiple histogram projections, Global Journal of Computer Science and Technology, Vol. 13, pp. 11-16.  Audithan, S., & Chandrasekaran, RM., 2009, Document text extraction from document images using haar discrete wavelet transform, European Journal of Scientific Research, Vol. 36, pp. 502-512.  Azadboni, M.K., & Behrad, A., Text detection and character extraction in color images using FFT domain filtering and SVM classification, 6th International Symposium on Telecommunications, Tehran, Nov. 6-8, 2012, pp. 794-799.
 Barina, d., 2011, Gabor wavelets in image processing. Available: http://www.fit.vutbr.cz/research/pubs/index.php?file=%2Fpub %2F9598%2Fprispevepdf&Id=9598.  Chandrasekaran, R., Chandrasekaran, RM., & Natarajan, P., Text localization and extraction in images using mathematical morphology and SVM, IEEE International Conference On Advances in Engineering, Science And Management, Nagapattinam, Tamil Nadu, March 30-31, 2012, pp. 55- 60.  Dinh, T.N., Park, J., & Lee, G.S., Low-complexity text extraction in korean signboards for mobile applications, IEEE International Conference on Computer and Information Technology, Sydney, NSW, July 8-11, 2008, pp. 333-337.  Eidheim, O.C., Introduction to mathematical morphology. Available: https://www.idi.ntnu.no/emner/tdt4265/lectures/lecture3b.pdf  Fan, W., Sun, J., Katsuyama, Y., Hotta, Y., & Naoi, S., Text detection in images based on grayscale decomposition and stroke extraction, Chinese Conference on Pattern Recognition, Nanjing, Nov. 4-6, 2009, pp. 1-4.
 Ghai, D., & Jain, N., 2013, Comparison of various text extraction techniques for images- a review, International Journal of Graphics & Image Processing, Vol. 3, pp. 210-218.  Ham, Y.K., Kang, M.S., Chung, H.K., & Park, R.H., 1995, Recognition of Raised Characters for Automatic Classification of Rubber Tires, Optical Engineering, Vol. 34, pp. 102–109.  Haritaoglu, I., Scene text extraction and translation for handheld devices, Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Hawaii, 2001, pp. 408–413.  Hassanzadeh, S., & Pourghassem, H., Fast logo detection based on morphological features in document image, IEEE 7th International Colloquium on Signal Processing and its Applications, Penang, Mar 4- 6 , 2011, pp. 283-286.  Jain, A.K., & Zhong, Y., 1996, Page segmentation using texture analysis, Pattern Recognition, Vol. 29, pp. 743–770.  Jung, K., Kim, K.I., & Jain, A.K., 2004, Text information extraction in images and video: a survey, Pattern Recognition, Vol. 37, pp. 977 – 997.
 Kanungo, T., Netanyahu, N.S., & Wu, A.Y., 2002, An efficient k-means clustering algorithm: analysis and implementation, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 24, pp. 881-892.  Kim, D.S., & Chien, S.I., Automatic car license plate extraction using modified generalized symmetry transform and image warping, Proceedings of International Symposium on Industrial Electronics, Pusan, Jun 12-16, 2001, pp. 2022–2027.  Kim, H.K., 1996, Efficient automatic text location method and content based indexing and structuring of video database, Journal of Visual Communication and Image Representation, Vol. 7, pp. 336-344.  Kumar, M., Kim, Y.C., & Lee, G.S., Text detection using multilayer separation in real scene images, 10th IEEE International Conference on Computer and Information Technology, Bradford, June 29- July1, 2010, pp. 1413-1417.  Lai, A.N., & Lee, G.S., Binarization by local k-means clustering for Korean text extraction, IEEE Symposium on Signal Processing and Information Technology, Sarajevo, Dec 16-19, 2008, pp. 117-122.
 Nathiya, N., & Pradeepa, K., Optical character recognition for scene text detection, mining and recognition, IEEE International Conference on Computational Intelligence and Computing Research, Enathi, Dec. 26-28, 2013, pp. 1-4.  Raj, H., & Ghosh, R., Devanagari text extraction from natural scene images, International Conference on Advances in Computing, Communications and Informatics (ICACCI), IEEE, New Delhi, India, Sept. 24-27, 2014, pp. 513-517.  Roy, S., Roy, P. P., Shivakumara, P., Louloudis, G., & Tan, C. L., HMM based multi oriented text recognition in natural scene image, Second IAPR Asian Conference on Pattern Recognition, Naha, Nov.5-8, 2013, pp. 288- 292.  Seeri, S.V., Giraddi, S., & Prashant B.M., A novel approach for Kannada text extraction, Proceedings of the International Conference on Pattern Recognition, Informatics and Medical Engineering, Tamil Nadu, India, Mar 21-23, 2012, pp. 444-448.  Song, Y., Liu, A., Pang, L., Lin, S., Zhang, Y. & Tang, S., A novel image text extraction method based on k-means clustering, Seventh IEEE/ACIS International Conference on Computer and Information Science, Portland, May 14-16, 2008, pp. 185-190.  Su, B., Lu, S., Phan, T.Q., & Tan, C.L., Character extraction in web image for text recognition, 21st International Conference on Pattern Recognition, Japan, Nov. 11-15, 2012, , pp. 3042-3045.
 Suen, C.Y., Lam, L., Guillevic, D., Strathy, N.W., 1996, Cheriet, M., Said, J.N. & Fan, R., Bank check processing system, International Journal of Imaging Systems and Technology, Vol. 7, pp. 392-403.  Watanabe, Y., Okada, Y., Kim, Y.B., & Takeda, T., Translation camera, Proceedings of International Conference on Pattern Recognition, Brisbane, Australia, 1998, pp. 613–617.  Yan, J., & Gao, X., 2014, Detection and recognition of text superimposed in images base on layered method, Neurocomputing, Vol. 134, pp. 3–14.  Yang, F., Qiu, Q., Bishop, M., & Wu, Q., Tag-assisted sentence confabulation for intelligent text recognition, IEEE Symposium on Computational Intelligence for Security and Defence Applications (CISDA), Ottawa, July 11-13, 2012, pp. 1-7.  Yao, J.L., Wang, Y.Q., Weng, L.B., & Yang, Y.P., Locating text based on connected component and SVM, International Conference on Wavelet Analysis and Pattern Recognition, Beijing, Nov 2-4, 2007, pp. 1418 - 1423.  Ye, Q., Huang, Q., Gao, W., & Zhao, D., 2005, Fast and robust text detection in images and video frames, Image and Vision Computing, Vol. 23, pp. 565–576.  Zhou, J., Lopresti, D., & Lei, Z., OCR for World Wide Web images, Proceedings of SPIE on Document Recognition IV, April 3, 1997, pp. 58–66.
 Zhou, J., Lopresti, D., & Tasdizen, T., Finding text in color images, Proceedings of SPIE on Document Recognition V, 1998, pp. 130–140.  Zaravi, D., Rostami, H., Malahzaheh, A., & Mortazavi, S.S., 2011, Text extraction using wavelet thresholding and new projection profile, World Academy of Science, Engineering and Technology, Vol. 5, pp. 528-531.  Zhang, Y., Wang, C., Xiao, B., & Shi, C., A new text extraction method incorporating local information, International Conference on Frontiers in Handwriting Recognition, Bari, Sept. 18-20, 2012, pp. 252-255.
Texture features based text extraction from images using DWT and K-means clustering

More Related Content

PPTX
Text extraction From Digital image
Kaushik Godhani
 
PPTX
Text extraction from images
Garby Baby
 
PPTX
Handwriting Recognition Using Deep Learning and Computer Version
Naiyan Noor
 
PPTX
Handwritten character recognition using artificial neural network
Harshana Madusanka Jayamaha
 
PPTX
Image to text Converter
Dhiraj Raj
 
DOCX
Digit recognition using mnist database
btandale
 
PDF
Image segmentation with deep learning
Antonio Rueda-Toicen
 
PDF
Facial Expression Recognition
Rupinder Saini
 
Text extraction From Digital image
Kaushik Godhani
 
Text extraction from images
Garby Baby
 
Handwriting Recognition Using Deep Learning and Computer Version
Naiyan Noor
 
Handwritten character recognition using artificial neural network
Harshana Madusanka Jayamaha
 
Image to text Converter
Dhiraj Raj
 
Digit recognition using mnist database
btandale
 
Image segmentation with deep learning
Antonio Rueda-Toicen
 
Facial Expression Recognition
Rupinder Saini
 

What's hot (20)

PPTX
Handwriting Recognition
Bindu Karki
 
PPTX
Convolutional Neural Networks
milad abbasi
 
PPTX
Face Detection Attendance System By Arjun Sharma
Arjun Agnihotri
 
PDF
Handwritten Character Recognition: A Comprehensive Review on Geometrical Anal...
iosrjce
 
PDF
CHARACTER RECOGNITION USING NEURAL NETWORK WITHOUT FEATURE EXTRACTION FOR KAN...
Editor IJMTER
 
PPTX
Cnn
Nirthika Rajendran
 
PDF
Image segmentation
Kuppusamy P
 
PPTX
Features image processing and Extaction
Ali A Jalil
 
PPTX
Unit3 dip
Imran Khan
 
PDF
Modern Convolutional Neural Network techniques for image segmentation
Gioele Ciaparrone
 
PPTX
Halftoning in Computer Graphics
University of Potsdam
 
DOCX
Hand Written Character Recognition Using Neural Networks
Chiranjeevi Adi
 
PDF
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
Jinwon Lee
 
PDF
Mask R-CNN
Chanuk Lim
 
PPTX
Segments in Graphics
Rajani Thite
 
PPTX
Handwritten Digit Recognition and performance of various modelsation[autosaved]
SubhradeepMaji
 
PPTX
Text detection and recognition from natural scenes
hemanthmcqueen
 
PPTX
Color Models
Mustafa Salam
 
PDF
Human Emotion Recognition
Chaitanya Maddala
 
PPTX
Convolutional Neural Networks
Ashray Bhandare
 
Handwriting Recognition
Bindu Karki
 
Convolutional Neural Networks
milad abbasi
 
Face Detection Attendance System By Arjun Sharma
Arjun Agnihotri
 
Handwritten Character Recognition: A Comprehensive Review on Geometrical Anal...
iosrjce
 
CHARACTER RECOGNITION USING NEURAL NETWORK WITHOUT FEATURE EXTRACTION FOR KAN...
Editor IJMTER
 
Cnn
Nirthika Rajendran
 
Image segmentation
Kuppusamy P
 
Features image processing and Extaction
Ali A Jalil
 
Unit3 dip
Imran Khan
 
Modern Convolutional Neural Network techniques for image segmentation
Gioele Ciaparrone
 
Halftoning in Computer Graphics
University of Potsdam
 
Hand Written Character Recognition Using Neural Networks
Chiranjeevi Adi
 
PR-231: A Simple Framework for Contrastive Learning of Visual Representations
Jinwon Lee
 
Mask R-CNN
Chanuk Lim
 
Segments in Graphics
Rajani Thite
 
Handwritten Digit Recognition and performance of various modelsation[autosaved]
SubhradeepMaji
 
Text detection and recognition from natural scenes
hemanthmcqueen
 
Color Models
Mustafa Salam
 
Human Emotion Recognition
Chaitanya Maddala
 
Convolutional Neural Networks
Ashray Bhandare
 
Ad

Similar to Texture features based text extraction from images using DWT and K-means clustering (20)

PDF
C04741319
IOSR-JEN
 
PPTX
Text extraction from natural scene image, a survey
SOYEON KIM
 
PDF
Methodology for eliminating plain regions from captured images
IAESIJAI
 
PDF
IRJET-MText Extraction from Images using Convolutional Neural Network
IRJET Journal
 
PDF
E1803012329
IOSR Journals
 
PDF
Text Extraction System by Eliminating Non-Text Regions
IJCSIS Research Publications
 
PDF
Inpainting scheme for text in video a survey
eSAT Journals
 
PDF
Scene Text Detection of Curved Text Using Gradiant Vector Flow Method
IJTET Journal
 
PDF
IRJET- Real-Time Text Reader for English Language
IRJET Journal
 
PDF
Text Extraction of Colour Images using Mathematical Morphology & HAAR Transform
IOSR Journals
 
PDF
ROBUST TEXT DETECTION AND EXTRACTION IN NATURAL SCENE IMAGES USING CONDITIONA...
ijiert bestjournal
 
PDF
Analysis and Comparison of various Methods for Text Detection from Images usi...
rahulmonikasharma
 
PDF
Cc31331335
IJMER
 
PPTX
Extracting text form video_using_matlab
Pula Likhitha
 
PDF
A Survey On Thresholding Operators of Text Extraction In Videos
CSCJournals
 
PDF
A Survey On Thresholding Operators of Text Extraction In Videos
CSCJournals
 
PDF
Text Extraction from Image using Python
ijtsrd
 
PDF
Scene text recognition in mobile applications by character descriptor and str...
eSAT Journals
 
PPTX
Presen_Segmentation
Vikas Goyal
 
PDF
Text Detection and Recognition: A Review
IRJET Journal
 
C04741319
IOSR-JEN
 
Text extraction from natural scene image, a survey
SOYEON KIM
 
Methodology for eliminating plain regions from captured images
IAESIJAI
 
IRJET-MText Extraction from Images using Convolutional Neural Network
IRJET Journal
 
E1803012329
IOSR Journals
 
Text Extraction System by Eliminating Non-Text Regions
IJCSIS Research Publications
 
Inpainting scheme for text in video a survey
eSAT Journals
 
Scene Text Detection of Curved Text Using Gradiant Vector Flow Method
IJTET Journal
 
IRJET- Real-Time Text Reader for English Language
IRJET Journal
 
Text Extraction of Colour Images using Mathematical Morphology & HAAR Transform
IOSR Journals
 
ROBUST TEXT DETECTION AND EXTRACTION IN NATURAL SCENE IMAGES USING CONDITIONA...
ijiert bestjournal
 
Analysis and Comparison of various Methods for Text Detection from Images usi...
rahulmonikasharma
 
Cc31331335
IJMER
 
Extracting text form video_using_matlab
Pula Likhitha
 
A Survey On Thresholding Operators of Text Extraction In Videos
CSCJournals
 
A Survey On Thresholding Operators of Text Extraction In Videos
CSCJournals
 
Text Extraction from Image using Python
ijtsrd
 
Scene text recognition in mobile applications by character descriptor and str...
eSAT Journals
 
Presen_Segmentation
Vikas Goyal
 
Text Detection and Recognition: A Review
IRJET Journal
 
Ad

Recently uploaded (20)

PPTX
The THESIS FINAL-DEFENSE-PRESENTATION.pptx
hshakobe3
 
PDF
The Rise of Impact Investing- How to Align Profit with Purpose
Harsh Mishra
 
PPTX
Bharatiya Antariksh Hackathon 2025 Idea Submission PPT.pptx
abhinavmemories2026
 
PPTX
Computer network topology notes for revision
BatoolRawat
 
PDF
Launch Your Data Science Career in Kochi – 2025
sharafas2563
 
PPTX
Introduction to Basics of Ethical Hacking and Penetration Testing -Unit No. 1...
AshutoshDeshmukh33
 
PPTX
Moving the Public Sector (Government) to a Digital Adoption
PaulYoung221210
 
PPTX
Introduction to Firewall Analytics - Interfirewall and Transfirewall.pptx
kuthubussaman1
 
PPTX
Logistic Regression ml machine learning.pptx
abdullahcocindia
 
PPTX
05. PRACTICAL GUIDE TO MICROSOFT EXCEL.pptx
macomputermacomputer
 
PDF
TRAFFIC-MANAGEMENT-AND-ACCIDENT-INVESTIGATION-WITH-DRIVING-PDF-FILE.pdf
chouziyu14
 
PPTX
advance b rammar.pptxfdgdfgdfsgdfgsdgfdfgdfgsdfgdfgdfg
rohullahansari5
 
PPTX
Introduction-to-Cloud-ComputingFinal.pptx
testnet29393883
 
PDF
“Getting Started with Data Analytics Using R – Concepts, Tools & Case Studies”
Nusrat Gulbarga
 
PPT
Reliability_Chapter_ presentation 1221.5784
abobaker13
 
PPTX
Supervised vs unsupervised machine learning algorithms
agarwal18harsh08
 
PDF
Data Science Trends & Career Guide---ppt
jisajoy3061
 
PDF
22.Patil - Early prediction of Alzheimer’s disease using convolutional neural...
ngaviet5
 
PPT
Quality review (1)_presentation of this 21
abobaker13
 
PDF
"Python Programming for Geospatial Data Science." ...
institute of Geoinformatics and Earth Observation at PMAS ARID Agriculture University of Rawalpindi
 
The THESIS FINAL-DEFENSE-PRESENTATION.pptx
hshakobe3
 
The Rise of Impact Investing- How to Align Profit with Purpose
Harsh Mishra
 
Bharatiya Antariksh Hackathon 2025 Idea Submission PPT.pptx
abhinavmemories2026
 
Computer network topology notes for revision
BatoolRawat
 
Launch Your Data Science Career in Kochi – 2025
sharafas2563
 
Introduction to Basics of Ethical Hacking and Penetration Testing -Unit No. 1...
AshutoshDeshmukh33
 
Moving the Public Sector (Government) to a Digital Adoption
PaulYoung221210
 
Introduction to Firewall Analytics - Interfirewall and Transfirewall.pptx
kuthubussaman1
 
Logistic Regression ml machine learning.pptx
abdullahcocindia
 
05. PRACTICAL GUIDE TO MICROSOFT EXCEL.pptx
macomputermacomputer
 
TRAFFIC-MANAGEMENT-AND-ACCIDENT-INVESTIGATION-WITH-DRIVING-PDF-FILE.pdf
chouziyu14
 
advance b rammar.pptxfdgdfgdfsgdfgsdgfdfgdfgsdfgdfgdfg
rohullahansari5
 
Introduction-to-Cloud-ComputingFinal.pptx
testnet29393883
 
“Getting Started with Data Analytics Using R – Concepts, Tools & Case Studies”
Nusrat Gulbarga
 
Reliability_Chapter_ presentation 1221.5784
abobaker13
 
Supervised vs unsupervised machine learning algorithms
agarwal18harsh08
 
Data Science Trends & Career Guide---ppt
jisajoy3061
 
22.Patil - Early prediction of Alzheimer’s disease using convolutional neural...
ngaviet5
 
Quality review (1)_presentation of this 21
abobaker13
 
"Python Programming for Geospatial Data Science." ...
institute of Geoinformatics and Earth Observation at PMAS ARID Agriculture University of Rawalpindi
 

Texture features based text extraction from images using DWT and K-means clustering

  • 1. TEXTURE FEATURES BASED TEXT EXTRACTION AND RECOGNITION SYSTEM Guided By:- Dr. Neelu Jain Associate Professor PEC University of Technology Presented By:- Divya Gera 13207003 ME(Electronics)
  • 2. Contents  Introduction  Motivation  Objectives  Literature Survey  Proposed Methodology  Experimental Results  Conclusion  Future Scope  References
  • 4.  Text content contains high level of semantic information as compared to visual information.  It provides important contents for information indexing and retrieval, automatic annotation and structuring of images.  Text extraction involves detection, localization, binarization, extraction, enhancement and recognition of the text from the given image.  These text characters are difficult to be detected and recognized due to their deviation of size, font, style, orientation, alignment, contrast, complex colored, textured background
  • 5. Types of text images i.) Document Images
  • 8. Process of Text Extraction & Recognition
  • 9. Applications of text extraction and recognition  Document indexing and retrieving Vehicle license plate detection
  • 10. Identification of parts in industrial automation Postal code from address on the envelope
  • 11. Street signs Libraries for computerized storage of books
  • 12. Bank cheque processing Blind and visually impaired person
  • 13. Text extraction techniques Classification of text extraction techniques
  • 14.  Region-based method uses the properties of the color or gray scale in the text region or their differences to the corresponding properties of the background.  Edge based method is focused on high contrast between the text and the background.  CC-based methods use a bottom-up approach by grouping small components into successively larger components until all regions are identified in the image.  Texture based method uses the fact that text in images has discrete textural properties that distinguish them from the background.
  • 15. Motivation  Text in multimedia documents or images contains important information for visual content understanding and information retrieval.  It is very challenging to design a general-purpose text extraction system due to the variations in text due to font size, style, color, orientation, and alignment.  In case of scene text images problem of low contrast, blurring, reflection, uneven illumination shadow, and perspective distortion exists along with text variations.  Also, less work has been done on multilingual text extraction.
  • 16. Objectives  To design and develop an algorithm for multilingual text extraction based on DWT coefficients, texture features, and k- means clustering algorithm.  To overcome the problems like uneven illumination and shadow, blurring and scratching, multi color, and complex background in the scene text extraction.  Develop a text recognition system for English language.  Compare the performance of the system in terms of parameters like Detection Rate, Precision Rate, and Recall Rate with the existing systems.  Developing a GUI for proposed algorithm.
  • 18. Author (year) Technique used Images Parameter Remarks Yao et al. (2007) CC and Support Vector Machine (SVM) Complex background images PR=64% RR=60% Pixels of each character assumed to have similar color. Lai et al. (2008) Edge detection and K- means clustering Signboard Images Efficient for uneven illumination Song et al. (2008) Histogram Projection and color based K- means clustering Chinese text PR=77.05% RR=75.63% K=3 gives best performance
  • 19. Author (year) Technique used Images Parameter Remarks Dinh et al. (2008) Edge detection and Histogram Projection Signboard Texts Low complexity algorithm. Fan et al. (2009) Stroke features and connected component Caption text images PR=95.2% RR= 94.5% Color information is not fully used Audithan et al. (2009) Haar DWT, logical AND operator, Dynamic thresholding Document images DR =94.8 % Independent of contrast. Angadi et al. (2010) Discrete Cosine Transform and texture features extraction Natural scene images DR=96.6% Inefficient for complex background
  • 20. Author (year) Technique used Images Parameter Remarks Anoual et al. (2010) Edge detection, texture features, connected compo- nent analysis Complex back-ground images PR=95% RR=89% Robust and effective. Kumar et al. (2010) CC Analysis ICDAR scene images PR=90% RR=89% Multilingual Text extraction. Hassanz- adeh et al. (2011) Morphological operator, Decision classifier Logo in document images PR=95.6% Accuracy=86. 9% A novel and fast method for logo detection. Chandra- sekaran et al. (2012) Morphological operation, and SVM ICDAR 2003 dataset PR=95% RR=92% Fails in case of high illumination
  • 21. Author (year) Technique used Images Parameter Remarks Zaravi et al. (2011) DWT, Dynamic thresholding, Region of Interest (ROI) Colored books and journal covers DR=91.20% Robust to noise. Zhang et al. (2012) Edge Enhancement and CC Web and caption text images DR=92.4% Insensitive to various types of background noises. Seeri et al. (2012) Median filter, Sobel edge detector, connected component labeling, order static filter. Kannada text images PR=84.21% RR=83.16% Accuracy = 75.77% Fails to extract very small characters.
  • 22. Author (Year) Technique used Images Parameter Remarks Azadboni et al. (2012) FFT Domain Filtering , SVM Classification, K-means clustering Scene text images DR= 98.10% Characters having uniform colour. Anupama et al. (2013) Morphology operators, Histogram Projection ( X and Y histogram) Handwritten Telugu document images. DR=98.54% Accuracy =98.29% Fails in case of touching characters & overlapping lines. Raj et al. (2014) CC based Natural Scene Images (Devanagari) PR= 72.8% RR=74.2 % Fails for small slanted/ curved text.
  • 24. Block diagram of text extraction
  • 25. Pre-processing  Although the color component may differ in a text region, yet it does not provide any information for text extraction.  Also, the processing of three components in the RGB image is difficult.  Colored input image is converted into gray scale image. Input image Gray Scale image
  • 26. 2D-DWT  The level-1 2D DWT wavelet has been applied decomposition to gray-scale images. It decomposes the image into four sub- bands: 1 approximation sub band and 3 detailed sub bands.  LL sub-band: Horizontal and vertical directions both are at low frequencies.  LH sub-band: Horizontal direction is at low frequency and vertical direction is at high frequency.  HL sub-band: Horizontal direction is at high frequency and vertical direction is at low frequency.  HH sub-band: Horizontal and vertical directions both are at high frequencies.
  • 27. 2D- DWT sub-bands (a) Approximation image (b) Vertical edges (c)Horizontal edges (d) Diagonal edges
  • 28. Feature extraction through sliding window  A small overlapped sliding window ( m×n) is scanned over each high frequency sub bands.  Zero padding is done, if required.  The text area has irregular texture property to a certain extent, so the text area can be looked as the special texture.  Language independent statistical features i.e. mean and standard deviation are calculated for the high frequency sub- bands.
  • 29. K-means clustering algorithm  Unsupervised technique of classification.  Divides the set of points into k clusters so that the intracluster similarity is high but intercluster similarity is low.  This is done by minimizing the sum of distances (euclidean distance) between the points and the cluster centers.  The clustering of image is done on the basis of texture features of LH, HL, and HH sub-bands.  This algorithm divides the image into k=3 clusters i.e. simple background, complex background and text clusters. The cluster that has the higher mean and standard deviation values is the text cluster.  For simple background, image is divided into 2 clusters i.e. background and text cluster.
  • 30. Morphological filter  The text cluster is mapped on a mask image by replacing the pixels in the text cluster by 1s and background cluster by 0s.  Morphological dilation operation is employed to fill the gap in the text region.  Dilation, basically, adds pixels to the boundaries of objects in an image. The number of pixels added to the objects in an image depends on the size and shape of the structuring element used to process the image.  In case of complex background, there may be some non text region in the mask image which is needed to be filter out.
  • 31. Steps involved in text extraction for simple background image (a) Input image, (b) Gray scale image, (c) Output of 2D DWT, (d) Background cluster, (e) text cluster, (f) Extracted text image. (c)(a) (b) (d) (e) (f)
  • 32. Steps involved in text extraction for complex background image (a) Input image (b) Gray scale image (c) Output of 2D DWT (d) Simple background cluster
  • 33. (e) Complex background cluster (f) Text cluster (g) Extracted text output
  • 34. Block diagram of text recognition
  • 35.  A template file has been created in which the letters A-Z, a-z, and numbers 0-9 are stored in the form of images.  A horizontal projection profile technique is used to isolate each line of the text.  The lines are segmented in the words by again scanning the image horizontally and detecting the spaces in between the words.  Each line is scanned vertically to detect and isolate each character within the line.  Then the correlation of each character in the target image with the template file character is found.
  • 37. GUI of the proposed technique (a) Input image
  • 42. Text extracted from caption text images (a) multicolor text (b) text of different size
  • 43. Text extracted from outdoor scene images
  • 44. Extracted text output from indoor scene images
  • 45. Text extracted from scene images with complex background
  • 46. Text extracted from image with curved text Text extracted from image with tilted text
  • 47. Text extraction in Punjabi (a) Multicolor document image (b) Caption text image
  • 48. Text extracted from multilingual images
  • 49. Parameters for performance evaluation of text extraction  Detection Rate (DR)= Ratio of the text regions correctly detected by the algorithm to the total number of text regions.  Precision Rate (PR)= Ratio of correctly detected characters to the sum of correctly detected characters plus false positives.  Recall Rate (RR)= Ratio of the correctly detected characters to sum of correctly detected characters plus false negatives.
  • 50. Dataset DR PR RR ICDAR and Kaist 98.93% 97.69% 97.83% Own dataset 98.99% 97.91% 98.27% Wavelets DR PR RR Haar 0.9979 0.9850 0.9835 db2 0.9942 0.9653 0.9497 bior1.3 0.9891 0.9618 0.8876 sym3 0.9894 0.9595 0.9131 coif1 0.9873 0.9364 0.9070 Comparative analysis of different wavelets Performance analysis of proposed text extraction technique
  • 51. Authors Method DR PR RR Angadi et al., 2010 Discrete Cosine Transform and texture features extraction 96.6% - - Kumar et al., 2010 Connected Component Analysis - 90% 89% Azadboni et al., 2012 FFT Domain Filtering , SVM Classification, K-means clustering 98.10% - - Proposed Method DWT and K-means Clustering 98.96% 97.8% 98.05% Comparison of proposed text extraction technique with others
  • 52. Output of text extraction and recognition system Sr. No. Input Image Extracted text output Recognized Text 1. 2.
  • 53. Sr. No. Input Image Extracted text output Recognized Text 3. 4.
  • 54. Parameter for text recognition performance evaluation Total Characters Falsely detected characters Accuracy (%) Proposed method 1000 49 95.10 Accuracy of text recognition system is defined as the ratio of correctly recognized characters to the total number of characters Accuracy of proposed text recognition system
  • 55. Conclusion  Text is successfully extracted from complex color images i.e. document text images, scene text images, caption text images, multi-colored text images, newspaper images, book cover images etc.  The proposed technique is insensitive to font size, style, color and alignment etc.  Independent of language.  The proposed method show that this method gives promising results in terms of DR (98.96%), PR (97.8%), RR (98.05%), and accuracy (95.1%) as compared to other methods.
  • 56. Future scope  Even though the performance of proposed system is excellent, there is always scope for improvement.  The proposed text extraction technique is inefficient for very low contrast and high illumination images.  The proposed system along with text to speech converter will be helpful to blind people.  The text after extraction and recognition can be translated from one language to other and will help the people to translate the text written on sign boards, street names etc. to their native language.
  • 57. References  Angadi, S.A., & Kodabagi, M., Text region extraction from low resolution natural scene images using texture features, IEEE 2nd International Advance Computing Conference, Patiala, India, Feb 19-20, 2010, pp. 121-128.  Anoual, H., Aboutajdine, D., Ensias, S.E., & Enset, A.J., Features extraction for text detection and localization, 5th International Symposium on I/V Communication and Mobile Network, Rabat, Sept. 30- Oct.2, 2010, pp. 1-4.  Anupama, N., Rupa, C., & Reddy, E.S., 2013, Character segmentation for telugu image document using multiple histogram projections, Global Journal of Computer Science and Technology, Vol. 13, pp. 11-16.  Audithan, S., & Chandrasekaran, RM., 2009, Document text extraction from document images using haar discrete wavelet transform, European Journal of Scientific Research, Vol. 36, pp. 502-512.  Azadboni, M.K., & Behrad, A., Text detection and character extraction in color images using FFT domain filtering and SVM classification, 6th International Symposium on Telecommunications, Tehran, Nov. 6-8, 2012, pp. 794-799.
  • 58.  Barina, d., 2011, Gabor wavelets in image processing. Available: http://www.fit.vutbr.cz/research/pubs/index.php?file=%2Fpub %2F9598%2Fprispevepdf&Id=9598.  Chandrasekaran, R., Chandrasekaran, RM., & Natarajan, P., Text localization and extraction in images using mathematical morphology and SVM, IEEE International Conference On Advances in Engineering, Science And Management, Nagapattinam, Tamil Nadu, March 30-31, 2012, pp. 55- 60.  Dinh, T.N., Park, J., & Lee, G.S., Low-complexity text extraction in korean signboards for mobile applications, IEEE International Conference on Computer and Information Technology, Sydney, NSW, July 8-11, 2008, pp. 333-337.  Eidheim, O.C., Introduction to mathematical morphology. Available: https://www.idi.ntnu.no/emner/tdt4265/lectures/lecture3b.pdf  Fan, W., Sun, J., Katsuyama, Y., Hotta, Y., & Naoi, S., Text detection in images based on grayscale decomposition and stroke extraction, Chinese Conference on Pattern Recognition, Nanjing, Nov. 4-6, 2009, pp. 1-4.
  • 59.  Ghai, D., & Jain, N., 2013, Comparison of various text extraction techniques for images- a review, International Journal of Graphics & Image Processing, Vol. 3, pp. 210-218.  Ham, Y.K., Kang, M.S., Chung, H.K., & Park, R.H., 1995, Recognition of Raised Characters for Automatic Classification of Rubber Tires, Optical Engineering, Vol. 34, pp. 102–109.  Haritaoglu, I., Scene text extraction and translation for handheld devices, Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Hawaii, 2001, pp. 408–413.  Hassanzadeh, S., & Pourghassem, H., Fast logo detection based on morphological features in document image, IEEE 7th International Colloquium on Signal Processing and its Applications, Penang, Mar 4- 6 , 2011, pp. 283-286.  Jain, A.K., & Zhong, Y., 1996, Page segmentation using texture analysis, Pattern Recognition, Vol. 29, pp. 743–770.  Jung, K., Kim, K.I., & Jain, A.K., 2004, Text information extraction in images and video: a survey, Pattern Recognition, Vol. 37, pp. 977 – 997.
  • 60.  Kanungo, T., Netanyahu, N.S., & Wu, A.Y., 2002, An efficient k-means clustering algorithm: analysis and implementation, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 24, pp. 881-892.  Kim, D.S., & Chien, S.I., Automatic car license plate extraction using modified generalized symmetry transform and image warping, Proceedings of International Symposium on Industrial Electronics, Pusan, Jun 12-16, 2001, pp. 2022–2027.  Kim, H.K., 1996, Efficient automatic text location method and content based indexing and structuring of video database, Journal of Visual Communication and Image Representation, Vol. 7, pp. 336-344.  Kumar, M., Kim, Y.C., & Lee, G.S., Text detection using multilayer separation in real scene images, 10th IEEE International Conference on Computer and Information Technology, Bradford, June 29- July1, 2010, pp. 1413-1417.  Lai, A.N., & Lee, G.S., Binarization by local k-means clustering for Korean text extraction, IEEE Symposium on Signal Processing and Information Technology, Sarajevo, Dec 16-19, 2008, pp. 117-122.
  • 61.  Nathiya, N., & Pradeepa, K., Optical character recognition for scene text detection, mining and recognition, IEEE International Conference on Computational Intelligence and Computing Research, Enathi, Dec. 26-28, 2013, pp. 1-4.  Raj, H., & Ghosh, R., Devanagari text extraction from natural scene images, International Conference on Advances in Computing, Communications and Informatics (ICACCI), IEEE, New Delhi, India, Sept. 24-27, 2014, pp. 513-517.  Roy, S., Roy, P. P., Shivakumara, P., Louloudis, G., & Tan, C. L., HMM based multi oriented text recognition in natural scene image, Second IAPR Asian Conference on Pattern Recognition, Naha, Nov.5-8, 2013, pp. 288- 292.  Seeri, S.V., Giraddi, S., & Prashant B.M., A novel approach for Kannada text extraction, Proceedings of the International Conference on Pattern Recognition, Informatics and Medical Engineering, Tamil Nadu, India, Mar 21-23, 2012, pp. 444-448.  Song, Y., Liu, A., Pang, L., Lin, S., Zhang, Y. & Tang, S., A novel image text extraction method based on k-means clustering, Seventh IEEE/ACIS International Conference on Computer and Information Science, Portland, May 14-16, 2008, pp. 185-190.  Su, B., Lu, S., Phan, T.Q., & Tan, C.L., Character extraction in web image for text recognition, 21st International Conference on Pattern Recognition, Japan, Nov. 11-15, 2012, , pp. 3042-3045.
  • 62.  Suen, C.Y., Lam, L., Guillevic, D., Strathy, N.W., 1996, Cheriet, M., Said, J.N. & Fan, R., Bank check processing system, International Journal of Imaging Systems and Technology, Vol. 7, pp. 392-403.  Watanabe, Y., Okada, Y., Kim, Y.B., & Takeda, T., Translation camera, Proceedings of International Conference on Pattern Recognition, Brisbane, Australia, 1998, pp. 613–617.  Yan, J., & Gao, X., 2014, Detection and recognition of text superimposed in images base on layered method, Neurocomputing, Vol. 134, pp. 3–14.  Yang, F., Qiu, Q., Bishop, M., & Wu, Q., Tag-assisted sentence confabulation for intelligent text recognition, IEEE Symposium on Computational Intelligence for Security and Defence Applications (CISDA), Ottawa, July 11-13, 2012, pp. 1-7.  Yao, J.L., Wang, Y.Q., Weng, L.B., & Yang, Y.P., Locating text based on connected component and SVM, International Conference on Wavelet Analysis and Pattern Recognition, Beijing, Nov 2-4, 2007, pp. 1418 - 1423.  Ye, Q., Huang, Q., Gao, W., & Zhao, D., 2005, Fast and robust text detection in images and video frames, Image and Vision Computing, Vol. 23, pp. 565–576.  Zhou, J., Lopresti, D., & Lei, Z., OCR for World Wide Web images, Proceedings of SPIE on Document Recognition IV, April 3, 1997, pp. 58–66.
  • 63.  Zhou, J., Lopresti, D., & Tasdizen, T., Finding text in color images, Proceedings of SPIE on Document Recognition V, 1998, pp. 130–140.  Zaravi, D., Rostami, H., Malahzaheh, A., & Mortazavi, S.S., 2011, Text extraction using wavelet thresholding and new projection profile, World Academy of Science, Engineering and Technology, Vol. 5, pp. 528-531.  Zhang, Y., Wang, C., Xiao, B., & Shi, C., A new text extraction method incorporating local information, International Conference on Frontiers in Handwriting Recognition, Bari, Sept. 18-20, 2012, pp. 252-255.