IJSRD - International Journal for Scientific Research & Development| Vol.

1, Issue 4, 2013 | ISSN (online): 2321-0613

Introduction to Wavelet Transform and Two Stage Image DE noising

Using Principal Component Analysis with Local Pixel Grouping
(LPGPCA) method

Mr. Ankur N. Shah1

1
PG Student (Computer Engineering Department)
1
C.U. Shah. College of Engineering & Technology, Wadhwan City, Gujarat.
1
[email protected]

Abstract— In past two decades there are various techniques II. INTRODUCTION TO WAVELET TRANSFORM
are developed to support variety of image processing (WT) AND PROBLEM WITH WT
applications. The applications of image processing include The wavelet transformation (WT) is combination of
medical, satellite, space, transmission and storage, radar and transforms that satisfy specific conditions. So wavelet
sonar etc. But noise in image effect all applications. So it is transform is a transform that have basis functions that are
necessary to remove noise from image. There are various shifted and expended version themselves. WT contains
methods and techniques are there to remove noise from information about both frequency information and spatial
images. Wavelet transform (WT) has been proved to be information. WT uses high pass filters, low pass filters and
effective in noise removal but this have some problems that Fourier transformations (FT). To satisfy condition for WT
is overcome by PCA method. This paper presents an the filter must be perfect Reconstruction filter which means
efficient image de-noising scheme by using principal that any distortion introduces by forward transform will be
component analysis (PCA) with local pixel grouping (LPG). cancel in the inverse transform. Examples of this type of
This method provides better preservation of image local filters are quadrature mirror filters. The WT breaks an image
structures. In this method a pixel and its nearest neighbors down into four subsamples or decimated images. They are
are modeled as a vector variable whose training samples are subsample by keeping every other pixel. The result consists
selected from the local window by using block matching of one image that has high pass filtered in both the
based LPG. In image de-noising, a compromise has to be horizontal and vertical directions. One that is high pass filter
found between noise reduction and preserving significant in horizontal and low pass filter in vertical. One that is high
image details. PCA is a statistical technique for simplifying pass filter in vertical and low pass filter in horizontal. One
a dataset by reducing datasets to lower dimensions. It is a that has low pass filtered in both directions. This transform
standard technique commonly used for data reduction in is typically implemented in the spatial domain by using 1-D
statistical pattern recognition and signal processing. This convolution filters. In order to perform the wavelet
paper proposes a de-noising technique by using a new transform using convolution filters a special type of
statistical approach, principal component analysis with local convolution called circular convolution must be used.
pixel grouping (LPG). This procedure is iterated second Circular convolution is performed by taking the underlying
time to further improve the de-noising performance, and the image array and extending it in a periodic manner to match
noise level is adaptively adjusted in the second stage. with symmetry implied by the discrete Fourier transform.
The convolution process start with origin of image and
Keywords: FT, WT, PCA, LPG, Noise, Pixel, De-noising. convolution masked aligned, So that the first value contains
contributions from the previous copy of periodic image.
I. INTRODUCTION This is important because we may want to perform the
Noise is any undesired information that spoil image. In wavelet transform on small blocks and eliminating outer
digital image noise arise during acquisition and/or row(s) and column(s) is not practical. There are various
transmission process. Following are different types of noise. filters can be used to implement WT. The two most
[4] commonly used are Daubechies and Haar. [1]
1) Gaussian/Normal noise Although WT has demonstrated its efficiency in de-noising
2) Rayleigh noise it uses fixed wavelet basis (with dilation and translation) to
3) Erlang/Gamma noise represent the image. For natural images however there is a
4) Exponential noise rich amount of different local structural patterns which
5) Impulse/Salt and pepper noise cannot be well represented by using only one fixed wavelet
6) Uniform noise basis. So WT based method can introduce many visual
7) Periodic noise artifacts in the de-noising output. To overcome this problem
De-noising is an essential step to improve the image quality. PCA based de-noising schema is used. [3]
There are different types of filters proposed for de-noising
like smoothing filters, frequency domain filters, nonlinear
filters etc. Wavelet transform (WT) has been proved to be
effective in noise removal. [7]

All rights reserved by www.ijsrd.com 880

 Introduction to Wavelet Transform and Two Stage Image DE noising Using Principal Component Analysis with Local Pixel Grouping (LPGPCA) method
(IJSRD/Vol. 1/Issue 4/2013/0026)

III. INTRODUCTION TO PRINCIPAL COMPONENT

ANALYSIS (PCA) AND LOCAL PIXEL GROUPING
(LPG)
The central idea of principal component analysis (PCA) is to
reduce the dimensionality of a data set consisting large
number of interrelated variables while retaining as much as
possible of the variation present in data set. This is achieved
by transforming to a new set of variables the principal Fig. 2: Flowchart for proposed two stage LPGPCA de-
components (PCs), which are uncorrelated and which are noising scheme
ordered so that the first few retain most of the variation A. ALGORITHM
present in all of the original variables. [2]
Principal component analysis (PCA) is a mathematical Following are steps.
procedure that uses an orthogonal transformation to convert [1] First step is to take noisy image.
a set of observations of possibly correlated variables into a [2] Second step is to local pixel grouping.
set of values of uncorrelated variables called principal [3] Third step is to PCA transform and de-noising.
components. [6] The number of principal components is less [4] Fourth step is to Inverse PCA transform. Calculate
than or equal to the number of original variables. This PSNR, SSIM, and COC.
transformation is defined in such a way that the first [5] Fifth step is update noise level & local pixel grouping.
principal component has as high a variance as possible and [6] Sixth step is to PCA transform and de-noising.
each succeeding component in turn has the highest variance [7] Seventh step is to Inverse PCA transform. Calculate
possible under the constraint that it be orthogonal to the PSNR, SSIM, and COC.
preceding components. Principal components are guaranteed [8] Eighth step is to obtain de-noised image.
to be independent only if the data set is jointly normally
distributed. PCA is sensitive to the relative scaling of the V. EXPERIMENT RESULTS
original variables. Depending on the field of application, it In proposed algorithm most of the computational cost
is also named the discrete Karhunen– Loève transform spends on LPG grouping and PCA transformation and thus
(KLT), the Hotelling transform or proper orthogonal complexity mainly depends on two parameters: the size of k
decomposition (POD). PCA can be done by eigenvalue of variable block and size L of training block. The proposed
decomposition of a data covariance matrix or singular value LPGPCA algorithm can be viewed as a completion and
decomposition of a data matrix, usually after mean centering extension of the PCA based de-noising algorithm.
the data for each attribute. [5] Following table – I show experiment values for different
Following fig-1 illustration of the modeling of black-white images and table – II shows color images.
LPGPCA based de-noising. [3] [9] Here we have experiment on different types of images like
TIF, JPG, GIF, BMP, PNG etc. The PSNR (Peak Signal to
Noise Ratio) use to measure the intensity difference between
two images. Generally quality of image can be measured by
PSNR. [8] SSIM(Structural Similarity Index) use to
measure image visual quality assessments. It shows
structural similarity between the target image and reference
image. COC shows the value for coefficient of correlation.

Fig. 1: Illustration of the modeling of LPGPCA based de-

noising
Grouping the training samples similar to the central KxK
block in the LxL training window is indeed a classification
problem and thus different grouping methods, such as block
matching, correlation-based matching, K-means
clustering, etc., can be employed based on different criteria.
Among them, the block matching method may be the
simplest yet very efficient one. It employ as LPG. [3]
Table (1): LPGPCA for black-white images
IV. FLOWCHART
Following fig-2 shows the flowchart for two stages LPG-

PCA de-noising scheme

All rights reserved by www.ijsrd.com 53

 Introduction to Wavelet Transform and Two Stage Image DE noising Using Principal Component Analysis with Local Pixel Grouping (LPGPCA) method
(IJSRD/Vol. 1/Issue 4/2013/0026)

Fig. 3: Comparison of first stage PSNR & second Stage

PSNR for black-white images
Table (2): LPGPCA for color images

Fig. 4: Comparison of first stage SSIM & second Stage

SSIM for black-white images Fig. 7: Comparison of first stage PSNR & second Stage
PSNR for color images

Fig. 5: Comparison of first stage COC & second Stage COC Fig. 8: Comparison of first stage SSIM & second Stage
for black-white images SSIM for color images

Fig 6: Comparison of first stage PSNR,SSIM, COC & Fig. 9: Comparison of first stage COC & second Stage COC
second Stage PSNR,SSIM, COC for black-white images for color images

All rights reserved by www.ijsrd.com 54

 Introduction to Wavelet Transform and Two Stage Image DE noising Using Principal Component Analysis with Local Pixel Grouping (LPGPCA) method
(IJSRD/Vol. 1/Issue 4/2013/0026)

REFERENCES
[1] Rafael C. Gonzalez and Richard E. Woods, “Digital
image processing”.
[2] I. T. Jolliffe, “Principal Component Analysis, Second
Edition”, Springer
[3] Lei Zhang, WeishengDong, DavidZhang, and
GuangmingShi “Two-stage image denoising by
principal component analysis with local pixel
grouping”, Pattern reorganization journal(2010).
[4] Mr. Ankur N. Shah, Dr. K.H.Wandra “Introduction to
noise, image restoration and comparison of various
methods of image restoration by removing noise from
Fig. 10: Comparison of first stage PSNR,SSIM, COC & image”, Indian journal of applied research, Oct 2012.
second Stage PSNR,SSIM, COC for color images [5] Mr. Zahid Alam “Image De-noising by Common Vector
Elimination in PCA (Principal Component Analysis)”,
International Journal of Scientific Engineering and
Technology, Volume No.1, Issue No.2 pg:22-26, April
2012.
[6] Vikas D Patil, Sachin D. Ruikar “PCA Based Image
Enhancement in Wavelet Domain”, International
Journal of Engineering Trends and Technology-
Volume3Issue1- 2012.
[7] K. John Peter, Dr K. Senthamarai Kannan, Dr S.
Arumugan, and G.Nagarajan “Two-stage image
denoising by Principal Component Analysis with Self
Similarity pixel Strategy”, IJCSNS International
Journal of Computer Science and Network Security,
VOL.11 No.5, May 2011.
[8] Y. Murali Mohan Babu, Dr. M.V. Subramanyam, and
Fig. 12: Banana image (color) various stage Dr. M.N. Giri Prasad “PCA based image denoising”,
Signal & Image Processing : An International Journal
(SIPIJ) Vol.3, No.2, April 2012.
[9] S.Kathiravan, N. Guhan Raja, R. Arun Kumar “An
investigation on image denoising technique using pixel-
component-analysis”, Innovative Systems Design and
Engineering,
[10] ISSN 2222-1727 (Paper) ISSN 2222-2871 (Online) Vol
2, No 4, 2011

Fig. 12: Boat image (black-white) various stage

VI. CONCLUSION
The LPGPCA technique given in this paper is useful with all
types of images like jpg, gif, tiff, bmp etc., it also work with
any size of image like 256x256 etc., But as size of image
increase it take more time because it have to work with
more number of pixels So further improvement can be done
in this direction. Additional this paper provides various
parameters to measure like PSNR, SSIM and COC.

All rights reserved by www.ijsrd.com 55