Enhancing readability of digital image using image processing - Full Report

ENHANCING READABILITY OF DIGITAL
IMAGE USING IMAGE PROCESSING
TECHNIQUES
A project report submitted in partial fulﬁllment of the requirements for
B.Tech. Project
B.Tech.
by
Akshay Kumar Lodha (2013IPG-007)
Sudesh Sindoskar (2013IPG-109)
Upendra Singh Sachan (2013IPG-118)
ABV INDIAN INSTITUTE OF INFORMATION
TECHNOLOGY AND MANAGEMENT
GWALIOR-474 010
2016

1
ABSTRACT
Smartphones have penetrated everyone’s lifestyle. With advance of technology, the
smartphones have become very powerful computers with very small form-factor. The
camera is an essential feature of every smartphone. We tend to digitize the printed
documents by clicking their photograph using the smartphone so that we can have a
digital backup for future reference or for ease of portability while reading on-the-go.
But this comes with added costs like:
(1) Reduced image quality results in reduced readability
(2) Noises like unwanted background keeps distracting the reader
(3) The uneven texture of photograph due to diﬀerent lighting condition like shadows,
etc.
As reading is an experience, readers want a noiseless environment. These things dis-
tracts users from reading and hence reduce their concentration and attention span. We
can use diﬀerent image processing techniques which can greatly increase the readabil-
ity of the document by removing/reducing noise and enhancing image.

TABLE OF CONTENTS
ABSTRACT 1
LIST OF FIGURES 3
1 INTRODUCTION AND LITERATURE SURVEY 5
1.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.1 Image Processing Techniques . . . . . . . . . . . . . . . . . . 6
1.1.1.1 Image Pre-processing . . . . . . . . . . . . . . . . . 7
1.1.1.2 Image Enhancement . . . . . . . . . . . . . . . . . 7
1.1.1.3 Image Segmentation . . . . . . . . . . . . . . . . . 7
1.1.1.4 Feature Extraction . . . . . . . . . . . . . . . . . . 7
1.1.1.5 Image Classiﬁcation . . . . . . . . . . . . . . . . . 7
1.1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 OpenCV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 DESIGN DETAILS AND IMPLEMENTATION 11
2.1 Detecting the edges . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.2 Canny Edge Detection . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Finding the contour . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Perspective Transformation . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Perspective Transformation Matrix . . . . . . . . . . . . . . . . 18
2.4 Image Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3 RESULTS AND DISCUSSION 25
3.1 RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 DISCUSSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4 CONCLUSION 34
2

TABLE OF CONTENTS 3
REFERENCES 34

LIST OF FIGURES
2.1 Basic Flow Diagram of Project . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Kernel = [1 x 1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Kernel = [3 x 3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Kernel = [5 x 5] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Kernel = [7 x 7] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Original Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Canny Edge Detection . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.8 Highlighting contour of document detected in the image . . . . . . . . 18
2.9 Image obtained after Perspective Trasformation . . . . . . . . . . . . . 21
2.10 Image enhanced using Histogram Equalization . . . . . . . . . . . . . 21
2.11 Image enhanced using Localized Histogram Equalization . . . . . . . . 22
2.12 Image enhanced using Adaptive Thresholding . . . . . . . . . . . . . . 22
2.13 Detailed flowchart of project . . . . . . . . . . . . . . . . . . . . . . . 23
2.14 Detailed flowchart of OpenCV code . . . . . . . . . . . . . . . . . . . 24
3.1 Result: Image Set 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Result: Document photographed from top . . . . . . . . . . . . . . . . 27
3.3 Result: Document photographed at an angle . . . . . . . . . . . . . . . 28
3.4 Result: Only the largest rectangle is selected . . . . . . . . . . . . . . . 29
3.5 Result: Image of Business Card . . . . . . . . . . . . . . . . . . . . . 30
3.6 Result: A coloured newspaper clipping . . . . . . . . . . . . . . . . . . 31
3.7 Result: Document with tables. Challenging example for finding contour 32
3.8 Result: Document with more than four edges is not detected . . . . . . 33
4

CHAPTER 1
INTRODUCTION AND
LITERATURE SURVEY
This chapter includes the details of image processing techniques, problem statement,
platform used to implement the project and literature reviews related to work done in
this field.
1.1 INTRODUCTION
One of the most important devices one can use to help their life is a smartphone. There
is so much that you can do with a smartphone and so many different ways in which they
play a key role in your life. It enhance and advance one’s personal life and working
life. With the advance of technology, the cost of processing speed is greatly reduced
and smartphones nowadays are nothing but a powerful full fledged computer with very
small form-factor. Now, everyone tend to work using their smartphone. So, everything
is going digital. But still, much of the information we access everyday is in printed
form. So, we digitize this by clicking the photograph of these documents and save
them in our smartphone for future reference or anytime access or reading them on their
smartphones. But this comes with some drawbacks such as:
(a) Images clicked are dependent on the quality of camera of the smartphone and hence
reducing the image quality which results in reduced readability
(b) Noise like unwanted background keeps distracting the reader
(c) The shadows projected over the document, when we try to click photograph from
top, reduces readability
(d) Tilted photographs tend to put extra pressure on reader’s mind
5

CHAPTER 1. INTRODUCTION AND LITERATURE SURVEY 6
Reading is an experience. The external factors which effects the reading speed of a
reader are readability of document (presentation) and concentration. The noises like
unwanted background, difficult to read image, etc. distracts the reader which decreases
their concentration and attention span.
For making a document more readable and presentable, we can do the following:
(a) Remove unwanted background
(b) Transform perspective of document to Top View
(c) Enhance the image by making text more readable and background less visible.
1.1.1 Image Processing Techniques
The definition of image processing is to processing of a digital image, i.e. extracting
the noise and any kind of irregularities available in an image using the digital computer.
The irregularity or noise may creep into the digital image either during its formation or
during its transformation etc. For mathematical analysis, an image may be elucidated
as a 2D function f(x,y) here x and y are spatial (plane) coordinates, and the amplitude of
f at any set of coordinates (x, y) is used as the intensity or gray level of the digital image
at that point. When x, y, and the intensity values of f are all limited, discrete quantities,
we can call the image a digital image. It is very significant that a digital image is made
of a finite number of elements, each one of them has a different value and location.
These elements are called image elements or pixels. Pixel is the most frequently used
term to describe the elements of a digital imageImage processing (2016).
Numerous techniques have been introduced in Image Processing during the last four to
five decades. Most of them are developed for enhancing images obtained from mili-
tary investigation flights, unmanned space crafts and space research. Image Processing
techniques are favoured due to easy accessibility of powerful computers, graphics soft-
ware, large size memory devices, etc.
The different Image Processing techniques are:
(i) Image pre-processing
(ii) Image enhancement
(iii) Image segmentation
(iv) Feature extraction
(v) Image classification

1.1.1.1 Image Pre-processing
In image pre-processing, image data registered by sensors or cameras prevent errors
associated with geometry and brightness values of the pixels. These errors are improved
using suitable mathematical models which are either definite or statistical models.
1.1.1.2 Image Enhancement
Image enhancement is the improvisation of digital image by changing the pixel bright-
ness values to modify its visual impact. Image enhancement includes a collection of
techniques that are used to modify the visual appearance of a digital image, or to change
the image to a form which is better suited for human or machine interpretationWang
et al. (2004).
1.1.1.3 Image Segmentation
Segmentation is one of the main problems in image processing. Image segmentation is
the method that classifies a digital image into its integral parts or objects. The depth to
which this subdivision or classification is carried out lies on the problem being solved,
i.e., the segmentation should terminate when the objects on interest in an application
have been isolated e.g., in independent air-to-ground target accession, assume our in-
terest is to distinguish vehicles on a road, the first step is to segment the road from the
digital image and then to subdivide the objects of the road down to probable vehicles.
Image thresholding methods are used for image segmentation.
1.1.1.4 Feature Extraction
The feature extraction methods are developed to extract features in synthetic aperture
radar images. This method withdraws high-level features required in order to conduct
classification of targets. Features are those items which individually narrate a target,
such as dimension, location, structure, composition etc. Segmentation methods are
applied to extract the wanted object from the scene so that measurements can be made
on it eventually. Quantitative measurements of object feature permit classification and
description of the image.
1.1.1.5 Image Classification
Image classification is the marking of a pixel or a group of pixels on the basis of its
grey value. Image classification is frequently used method of information extraction.
In Classification, mostly multiple features are implemented in a set of pixels i.e., many
images of a specific object are required.

1.1.2 Problem Statement
To enhance the readability of a photographed printed document using image processing
techniques by removing noise, enhancing image and transforming perspective.
1.1.3 Motivation
The main focus of this project is to develop a system to increase readability of the digital
image of a document by using image processing and computer vision. As reading is all
about experience, the noise free document is a must for every reader.
1.2 OpenCV
We have used OpenCV using Python interface to implement our project. OpenCV
(Open Source Computer Vision) is a library of programming functions mainly aimed
at real-time computer vision. OpenCV | OpenCV (2016) OpenCV is aimed at providing
the tools needed to solve computer-vision problems. It contains a mix of low-level
image-processing functions and high-level algorithms such as face detection, pedestrian
detection, feature matching, and tracking. The library has been downloaded more than
3 million times. OpenCV (2016)
1.3 Literature Review
We mainly deal with the literatures about different image processing algorithms cover-
ing topics in Image Enhancement, Feature extraction and Image Segmentation.
-B. Chitradevi and P.Srimathi (2014) gives an overview on Image Processing Tech-
niques in their report. The preservation of original data precision, versatility and re-
peatability are thr principle advantages of Digital Image Processing. The various image
processing techniques are Image preprocessing, image enhancement, image segmenta-
tion, feature extraction and image classification. The image enhancement algorithms
are interactive and application dependent. Some image enhancement techniques are
Contrast stretching, Noise filtering, Histogram modification. Chitradevi and P.Srimathi
(1970)
-Bing Wang and ShaoSheng Fan (2009) proposed that traditional Canny Edge detection
algorithm has the defect that being vulnerable to various noise disturbances. It can be
greatly improved by replacing the pre-processing step Gaussian filter with self-adaptive
filter and further improved by morphological thinning to thin the edge. Wang and Fan

(2009)
The use of Gaussian filter in traditional algorithm posed a problem by not only smooth-
ing the noise but smoothing the edges also, so some edge is weakened. The difficulty
to determine the Dual-threshold can’t remove the noise or makes lose some edge.
-In the work by T.L.Tan, K.S. Sim and C.P.Tso (2012), a new approach of nonlin-
ear Histogram Equalization is presented, which is not only able to enhance the image
contrast but it also preserves the background brightness. Tan et al. (2012)
-In the work by Kaiming He, Jian Sun, Xiaoou Tang (2009), they have proposed a
simple but effective image prior called dark channel prior to remove the haze from a
single input image. The dark channel prior is a kind of statistics of the haze-free out-
door images based on a key observation that most local patches in haze-free outdoor
images contain some pixels which have very low intensities in at least one color chan-
nel. Using this prior with the haze imaging model, they directly estimated the thickness
of haze and recover high quality haze-free image. He et al. (2011)
-Michael Donoser and Horst Bischof (2006) proposed a novel concept for tracking of
Maximally Stable Extremal Regions (MSERs). The approach uses temporal informa-
tion to improve the computational time and detect stability of single MSER. This can
be widely used in tracking license plates, face tracking, segmenting the fiber network,
etc. Donoser and Bischof (2006)
-Per-Erik Forssen and David G.Lowe (2007) introduced an affine invariant shape de-
scriptor for maximally stable extremal regions. The descriptor is computed using the
scale invariant feature transform (SIFT), with re-sampled MSER binary mask as input.
This provides better robustness to illumination change and nearby occlusions than ex-
isting methods. Forssen and Lowe (2007)
-In the work by M.Cheriet, J.N. Said, C.Y. Suen (1998), the presented a general recur-
sive thresholding technique for image segmentation by extending Otsu’s method. This
approach of image segmentation has been implemented in the scope of document im-
ages, specifically real-life bank cheques. At each recursion, the new approach segments
first the object with lowest intensity from the given image and this process continues
until there is one object left in the image. Cheriet et al. (1998)
-In the work done by Pranay Yadav (2015), proposed a method for noise removal in
coloured image by modified adaptive threshold median filter. The proposed filter is
very effective for random valued impulse noise because practically noise is not uniform

over the channel. He used concept of both minimum and maximum thresholds to detect
the positive and negative noise. The threshold values are different for different noise
density. Yadav (2015)
- Md. Habibur Rahman and Md. Rafiqul Islam (2013) proposed a modified version
of the watershed algorithm for image segmentation. The overcome the problem of over
segmentation by introducing adaptive masking and a thresholding mechanism over each
color channel befor combining the segmentation from each channel into the final one.
The proposed modified watershed algorithm is also faster than many other segmenta-
tion algorithms, making it fit for real-time application. Rahman and Islam (2013)

CHAPTER 2
DESIGN DETAILS AND
IMPLEMENTATION
To begin, first we need to thoroughly analyse the properties of document that we need to
extract from the image. The typical properties and assumptions of image of a document
are:
(i) The image clicked is clear enough to read the contents of the document.
(ii) Generally, the content (text) is in black ink while the background (paper) is white
coloured.
(iii) The document has exactly four straight edges (and hence the vertices)
(iv) The tilt of perspective from the top view is not more than 45 degrees.
(v) Maximum portion of the image is covered by the document.
(vi) Nothing overlaps or obstructs the document in the image (like fingers, weights,
etc.).
The task is to enhance the readability of a photographed printed document using
image processing techniques by removing noise, enhancing image and transforming
perspective has broadly 4 essential steps:
(1) Detecting the edges of document
(2) Using edges to find contour of the document
(3) Applying perspective transformation to obtain bird’s eye / top-down view of the
document.
(4) Enhancing image quality to increase readability
11

CHAPTER 2. DESIGN DETAILS AND IMPLEMENTATION 12
Figure 2.1: Basic Flow Diagram of Project

2.1 Detecting the edges
In an image, an edge is a curve that follows a path of rapid change in image intensity.
Edges are often associated with the boundaries of objects in a scene. Edge detection is
used to identify the edges in an imageFeature Detectors - Sobel Edge Detector (2016).
There are plenty of edge detection techniques, such as Roberts, Prewitt, Sobel and
CannyPrewitt operator (2016) Realtime Computer Vision with OpenCV - ACM Queue
(2016).
According to PSNR, Sobel gives the most quality image while Roberts gives the
least quality imageSingh and Singh (2015). But, in comparison, Canny edge detection
goes a bit further by removing speckle noise with a low pass filter first, then applying
a Sobel filter, and then doing non-maximum suppression to pick out the best pixel for
edges when there are multiple possibilities in a local neighbourhood. So, we choose
Canny edge detection technique.
2.1.1 Pre-processing
Edge detection techniques are susceptible to noise. Camera is a man-made device so it
is not 100% perfect. The raw image i.e. the original image of the document contains
noise which give false results. So we apply a Gaussian filter on the image to smooth
out the noise. The size of the kernel must be chosen accordingly to so that an optimum
amount of noise is filtered out but the edge does not blurred.

Figure 2.2: Kernel = [1 x 1]

The optimum value kernel is chosen as (5,5) to avoid loss of details of edge and
remove the noise eﬀectively.

The Gaussian ﬁlter takes input of image with single layer, so we convert the RGB
image to Grayscale.
Code:
image = cv2 . imread ( args [ " image " ] )
gray = cv2 . cvtColor ( image , cv2 .COLOR_BGR2GRAY)
gray = cv2 . GaussianBlur ( gray , (5 , 5) , 0)
2.1.2 Canny Edge Detection
After the image is pre processed, we apply the Canny Edge Detection Wang and Fan
(2009) with minVal= 75 and maxVal=200
Code:
edged = cv2 . Canny ( gray , 75 , 200)
Figure 2.6: Original Image

Figure 2.7: Canny Edge Detection
2.2 Finding the contour
After identifying the edges, we find the contour of the document. As mentioned earlier,
we assumed that most of the part of the image is occupied by the document and it has
four edges.
We find the contours using cv2.findContours with contour retrieval mode as RETR_LIST.
We only need the contours; we drop the hierarchy. We then keep only the largest five
contours in descending order (assuming most of the part of the image is occupied by
the document) to speed the process by reducing number of iterations.
Now, we count the number of points in the contours (which should be equal to four
according to our assumption) in descending order. If number of points is equal to four,
then we assume that to be contour of our document. Next, we simply highlight the
estimated contour of the document on the original image to verify.
Code:
( cnts , _ ) = cv2 . findContours ( edged . copy ( ) , cv2 . RETR_LIST ,
cv2 . CHAIN_APPROX_SIMPLE)
c n t s = s o r t e d ( cnts , key = cv2 . contourArea , r e v e r s e = True ) [ : 5 ]
f o r c in c o n t o u r s s :
p e r i = cv2 . arcLength ( c , True )
approx = cv2 . approxPolyDP ( c , 0.02 ∗ peri , True )
i f len ( approx ) == 4:
document_contour = approx

break
cv2 . drawContours ( image , [ document_contour ] , −1, (0 , 255 , 0) , 2)
Figure 2.8: Highlighting contour of document detected in the image
2.3 Perspective Transformation
Now that we have identified the document edges, we now need to apply perspective
transformation to obtain a top view for better readability.
2.3.1 Perspective Transformation Matrix
To perform a perspective transformation, we need to first find out the Transformation
Matrix. For this, we name the four points as top-left, top-right, bottom-left and bottom-
right.
The coordinate top-left point is (0,0). We find the maximum height of the new
image by simply using distance formula (x2 − x1)2 + (y2 − y1)2 between the points

top-right and bottom-right or top-left and bottom-left, whichever is maximum. Simi-
larly, the maximum width of the new image will be distance between points top-right
and top-left or bottom-right and bottom-left, whichever is maximum. Now, we con-
struct a distance array with values: [0,0], [maxWidth -1,0],[maxWidth-1,maxHeight-
1],[0,maxHeight-1]
Using this array and the original coordinates of the document, find the Perspective
Transformation Matrix using code: M = cv2.getPerspectiveTransform(rect, distance),
where rect = [top-left, top-right, bottom-left,bottom-right]
The coordinates of the pixels of the new image, which we get after perspective
transformation, can be given as (u, v) corresponding to pixels of original image (x, y).
u =
a0x + a1y + a2
c0x + c1y + 1
v =
b0x + b1y + b2
c0x + c1y + 1
where: a0, a1, a2, b0, b1, b2, c0, c1 are constants.
Now, we need to find the value of these constants. To find these we have relations
of 4 pairs i.e. top-left, top=right, bottom-left and bottom-right.
X ∗ A = U
where:
X =


x0 y0 1 0 0 0 −x0u0 −y0u0
x1 y1 1 0 0 0 −x1u1 −y1u1
x2 y2 1 0 0 0 −x2u2 −y2u2
x3 y3 1 0 0 0 −x3u3 −y3u3
0 0 0 x0 y0 1 −x0v0 −y0v0
0 0 0 x1 y1 1 −x1v1 −y1v1
0 0 0 x2 y2 1 −x2v2 −y2v2
0 0 0 x3 y3 1 −x3v3 −y3v3


A =


a0
a1
a2
b0
b1
b2
c0
c1



U =


u0
u1
u2
u3
v0
v1
v2
v3


Now, using this Perspective Transformation Matrix we transform the image to top
view.
Code:
r e c t = o r d e r _ p o i n t s ( p t s )
( t l , tr , br , bl ) = r e c t
widthA = np . s q r t ( ( ( br [ 0 ] − bl [ 0 ] ) ∗∗ 2) + ( ( br [ 1 ] − bl [ 1 ] ) ∗∗ 2 ) )
widthB = np . s q r t ( ( ( t r [ 0 ] − t l [ 0 ] ) ∗∗ 2) + ( ( t r [ 1 ] − t l [ 1 ] ) ∗∗ 2 ) )
maxWidth = max ( i n t ( widthA ) , i n t ( widthB ) )
heightA = np . s q r t ( ( ( t r [ 0 ] − br [ 0 ] ) ∗∗ 2) + ( ( t r [ 1 ] − br [ 1 ] ) ∗∗ 2 ) )
heightB = np . s q r t ( ( ( t l [ 0 ] − bl [ 0 ] ) ∗∗ 2) + ( ( t l [ 1 ] − bl [ 1 ] ) ∗∗ 2 ) )
maxHeight = max ( i n t ( heightA ) , i n t ( heightB ) )
d i s t a n c e = np . a r r a y ( [
[0 , 0] ,
[ maxWidth − 1 , 0] ,
[ maxWidth − 1 , maxHeight − 1] ,
[0 , maxHeight − 1 ] ] , dtype = " f l o a t 3 2 " )
M = cv2 . g e t P e r s p e c t i v e T r a n s f o r m ( r ect , d i s t a n c e )
warped = cv2 . warpPerspective ( image , M, ( maxWidth , maxHeight ) )
2.4 Image Enhancement
The last step is image enhancement. The new cropped image with top-view is ready
but we can increase its readability by enhancing the image by making the background
of document(paper) even. Two of the most popular Image Enhancement methods are
Histogram EqualizationTan et al. (2012) and Adaptive ThresholdingPoint Operations -
Adaptive Thresholding (2016) Stark (2000).
Histogram Equalisation, although improves image, but when compared to Adaptive
Thresholding, the latter seems to doing a better job. So, we use adaptive thresholding
to enhance readability of the image of document.

Figure 2.9: Image obtained after Perspective Trasformation
Figure 2.10: Image enhanced using Histogram Equalization

Figure 2.11: Image enhanced using Localized Histogram Equalization
Figure 2.12: Image enhanced using Adaptive Thresholding

Figure 2.13: Detailed ﬂowchart of project

Figure 2.14: Detailed ﬂowchart of OpenCV code

CHAPTER 3
RESULTS AND DISCUSSION
3.1 RESULTS
The program output the enhanced image for better reading experience by removing the
unwanted background and clear and crisp texts.
3.2 DISCUSSION
Although the adaptive thresholding produces a good readable image but the informa-
tion about colour is lost. This can be hoped to solve by using haze removal algorithm
as the background of the page can be treated as haze which can be subtracted retaining
the colour informationHe et al. (2011).
The program also fails when four edges are not detected. It may pose problem when
document is of diﬀerent shape or an irregular rectangle. The irregular rectangle prob-
lem can be solved by changing the epsilon value of approxPolyDP function. Also
Hough Transform can also be used detecting linesHough transform - Wikipedia, the
free encyclopedia (2016).
25

CHAPTER 3. RESULTS AND DISCUSSION 26
(a) Original Image (b) Canny Edge Detection
(c) Contour Detection (d) Perspective Transformation
(e) Final Result
Figure 3.1: Result: Image Set 1

(e) Final Result
Figure 3.2: Result: Document photographed from top

(e) Final Result
Figure 3.3: Result: Document photographed at an angle

(e) Final Result
Figure 3.4: Result: Only the largest rectangle is selected

(e) Final Result
Figure 3.5: Result: Image of Business Card

(e) Final Result
Figure 3.6: Result: A coloured newspaper clipping

(e) Final Result
Figure 3.7: Result: Document with tables. Challenging example for ﬁnding contour

Figure 3.8: Result: Document with more than four edges is not detected

CHAPTER 4
CONCLUSION
The readability of digital images of documents (especially Black and White) can be
greatly increased by removing the noise and enhancing image using adaptive threshold.
In future, an algorithm for adaptive thresholding of coloured images can be used to
retain the colour information. The Perspective transformation can also be increased to
a great extent if angle of inclination (can be approximated on the basis of the diﬀerence
of edge length of two opposite sides) can is also taken into account. Further, a text
detecting algorithmXiao and Yan (2003) can be used to replace the printed letters with
computer generated text to take readability to next level.
34

REFERENCES
[1] Cheriet, M., Said, J. N. and Suen, C. Y.: 1998, A recursive thresholding technique
for image segmentation, IEEE Transactions on Image Processing 7(6), 918–921.
[2] Chitradevi, B. and P.Srimathi: 1970, An Overview on Image Processing Tech-
niques, International Journal of Innovative Research in Computer and Communi-
cation Engineering .
URL: http://www.rroij.com/abstract.php?abstractid = 47175
[3] Donoser, M. and Bischof, H.: 2006, Eﬃcient Maximally Stable Extremal Region
(MSER) Tracking, 2006 IEEE Computer Society Conference on Computer Vision
and Pattern Recognition (CVPR’06), Vol. 1, pp. 553–560.
[4] Feature Detectors - Sobel Edge Detector: 2016.
URL: http://homepages.inf.ed.ac.uk/rbf/HIPR2/sobel.htm
[5] Forssen, P. E. and Lowe, D. G.: 2007, Shape Descriptors for Maximally Stable
Extremal Regions, 2007 IEEE 11th International Conference on Computer Vision,
pp. 1–8.
[6] He, K., Sun, J. and Tang, X.: 2011, Single Image Haze Removal Using Dark
Channel Prior, IEEE Transactions on Pattern Analysis and Machine Intelligence
33(12), 2341–2353.
[7] Hough transform - Wikipedia, the free encyclopedia: 2016.
URL: https://en.wikipedia.org/wiki/Houghtransform
[8] Image processing: 2016. Page Version ID: 740950083.
URL: https://en.wikipedia.org/w/index.php?title=Imageprocessingoldid =
740950083
[9] OpenCV: 2016. Page Version ID: 739519893.
URL: https://en.wikipedia.org/w/index.php?title=OpenCVoldid=739519893
[10] OpenCV | OpenCV: 2016.
URL: http://opencv.org/
35

REFERENCES 36
[11] Point Operations - Adaptive Thresholding: 2016.
URL: http://homepages.inf.ed.ac.uk/rbf/HIPR2/adpthrsh.htm
[12] Prewitt operator: 2016. Page Version ID: 698509644.
URL: https://en.wikipedia.org/w/index.php?title=Prewitto peratoroldid =
698509644
[13] Rahman, M. H. and Islam, M. R.: 2013, Segmentation of color image using adap-
tive thresholding and masking with watershed algorithm, 2013 International Con-
ference on Informatics, Electronics Vision (ICIEV), pp. 1–6.
[14] Realtime Computer Vision with OpenCV - ACM Queue: 2016.
URL: http://queue.acm.org/detail.cfm?id=2206309
[15] Singh, S. and Singh, R.: 2015, Comparison of various edge detection techniques,
2015 2nd International Conference on Computing for Sustainable Global Devel-
opment (INDIACom), pp. 393–396.
[16] Stark, J. A.: 2000, Adaptive image contrast enhancement using generalizations of
histogram equalization, IEEE Transactions on Image Processing 9(5), 889–896.
[17] Tan, T. L., Sim, K. S. and Tso, C. P.: 2012, Image enhancement using background
brightness preserving histogram equalisation, Electronics Letters 48(3), 155–157.
[18] Wang, B. and Fan, S.: 2009, An Improved CANNY Edge Detection Algorithm,
Second International Workshop on Computer Science and Engineering, 2009.
WCSE ’09, Vol. 1, pp. 497–500.
[19] Wang, Z., Bovik, A. C., Sheikh, H. R. and Simoncelli, E. P.: 2004, Image quality
assessment: from error visibility to structural similarity, IEEE Transactions on
Image Processing 13(4), 600–612.
[20] Xiao, Y. and Yan, H.: 2003, Text region extraction in a document image based on
the Delaunay tessellation, Pattern Recognition 36(3), 799–809.
URL: http://www.sciencedirect.com/science/article/pii/S0031320302000821
[21] Yadav, P.: 2015, Color image noise removal by modified adaptive threshold me-
dian filter for RVIN, 2015 International Conference on Electronic Design, Com-
puter Networks Automated Verification (EDCAV), pp. 175–180.

Enhancing readability of digital image using image processing - Full Report

More Related Content

What's hot (19)

Similar to Enhancing readability of digital image using image processing - Full Report (20)

Recently uploaded (20)

Enhancing readability of digital image using image processing - Full Report