|Views: 688|Likes: 0

Published by ijcsis

The paper presents the novel hybrid wavelet transform generation technique using two orthogonal transforms. The orthogonal transforms are used for analysis of global properties of the data into frequency domain. For studying the local properties of the signal, the concept of wavelet transform is introduced, where the mother wavelet function gives the global properties of the signal and wavelet basis functions which are compressed versions of mother wavelet are used to study the local properties of the signal. In wavelets of some orthogonal transforms the global characteristics of the data are hauled out better and some orthogonal transforms might give the local characteristics in better way. The idea of hybrid wavelet transform comes in to picture in view of combining the traits of two different orthogonal transform wavelets to exploit the strengths of both the transform wavelets.

The paper proves the worth of hybrid wavelet transforms for the image compression which can further be extended to other image processing applications like steganography, biometric identification, content based image retrieval etc. Here the hybrid wavelet transforms are generated using four orthogonal transforms alias Discrete Cosine transform (DCT), Discrete Hartley transform (DHT), Discrete Walsh transform (DWT) and Discrete Kekre transform (DKT). The comparison of the hybrid wavelet transforms is also done with the original orthogonal transforms and their wavelet transforms. The experimentation results have shown that the transform wavelets have given better quality of image compression than the respective original orthogonal transforms but for hybrid transform wavelets the performance is best. Here the hybrid of DCT and DKT gives the best results among the combinations of the four mentioned image transforms used for generating hybrid wavelet transforms.

The paper proves the worth of hybrid wavelet transforms for the image compression which can further be extended to other image processing applications like steganography, biometric identification, content based image retrieval etc. Here the hybrid wavelet transforms are generated using four orthogonal transforms alias Discrete Cosine transform (DCT), Discrete Hartley transform (DHT), Discrete Walsh transform (DWT) and Discrete Kekre transform (DKT). The comparison of the hybrid wavelet transforms is also done with the original orthogonal transforms and their wavelet transforms. The experimentation results have shown that the transform wavelets have given better quality of image compression than the respective original orthogonal transforms but for hybrid transform wavelets the performance is best. Here the hybrid of DCT and DKT gives the best results among the combinations of the four mentioned image transforms used for generating hybrid wavelet transforms.

The paper presents the novel hybrid wavelet transform generation technique using two orthogonal transforms. The orthogonal transforms are used for analysis of global properties of the data into frequency domain. For studying the local properties of the signal, the concept of wavelet transform is introduced, where the mother wavelet function gives the global properties of the signal and wavelet basis functions which are compressed versions of mother wavelet are used to study the local properties of the signal. In wavelets of some orthogonal transforms the global characteristics of the data are hauled out better and some orthogonal transforms might give the local characteristics in better way. The idea of hybrid wavelet transform comes in to picture in view of combining the traits of two different orthogonal transform wavelets to exploit the strengths of both the transform wavelets.

The paper proves the worth of hybrid wavelet transforms for the image compression which can further be extended to other image processing applications like steganography, biometric identification, content based image retrieval etc. Here the hybrid wavelet transforms are generated using four orthogonal transforms alias Discrete Cosine transform (DCT), Discrete Hartley transform (DHT), Discrete Walsh transform (DWT) and Discrete Kekre transform (DKT). The comparison of the hybrid wavelet transforms is also done with the original orthogonal transforms and their wavelet transforms. The experimentation results have shown that the transform wavelets have given better quality of image compression than the respective original orthogonal transforms but for hybrid transform wavelets the performance is best. Here the hybrid of DCT and DKT gives the best results among the combinations of the four mentioned image transforms used for generating hybrid wavelet transforms.

The paper proves the worth of hybrid wavelet transforms for the image compression which can further be extended to other image processing applications like steganography, biometric identification, content based image retrieval etc. Here the hybrid wavelet transforms are generated using four orthogonal transforms alias Discrete Cosine transform (DCT), Discrete Hartley transform (DHT), Discrete Walsh transform (DWT) and Discrete Kekre transform (DKT). The comparison of the hybrid wavelet transforms is also done with the original orthogonal transforms and their wavelet transforms. The experimentation results have shown that the transform wavelets have given better quality of image compression than the respective original orthogonal transforms but for hybrid transform wavelets the performance is best. Here the hybrid of DCT and DKT gives the best results among the combinations of the four mentioned image transforms used for generating hybrid wavelet transforms.

See More

See less

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, 2011

Inception of Hybrid Wavelet Transform using TwoOrthogonal Transforms and It’s use for ImageCompression

Dr. H.B.Kekre

,Senior Professor,Computer Engineering Department,SVKM’s NMIMS (Deemed-to-beUniversity)Vile Parle(W), Mumbai, India.hbkekre@yahoo.com,

Dr.Tanuja K. Sarode

Assistant Professor Computer EngineeringDepartment,Thadomal Shahani EngineeringCollege, Bandra(W), Mumbai, India.tanuja_0123@yahoo.com

Sudeep D. Thepade

Associate Professor Computer Engineering Department,SVKM’s NMIMS (Deemed-to-beUniversity)Vile Parle(W), Mumbai, India.sudeepthepade@gmail.com

Abstract

—The paper presents the novel hybrid wavelet transformgeneration technique using two orthogonal transforms. Theorthogonal transforms are used for analysis of global propertiesof the data into frequency domain. For studying the localproperties of the signal, the concept of wavelet transform isintroduced, where the mother wavelet function gives the globalproperties of the signal and wavelet basis functions which arecompressed versions of mother wavelet are used to study the localproperties of the signal. In wavelets of some orthogonaltransforms the global characteristics of the data are hauled outbetter and some orthogonal transforms might give the localcharacteristics in better way. The idea of hybrid wavelettransform comes in to picture in view of combining the traits of two different orthogonal transform wavelets to exploit thestrengths of both the transform wavelets.The paper proves the worth of hybrid wavelettransforms for the image compression which can further beextended to other image processing applications likesteganography, biometric identification, content based imageretrieval etc. Here the hybrid wavelet transforms are generatedusing four orthogonal transforms alias Discrete Cosine transform(DCT), Discrete Hartley transform (DHT), Discrete Walshtransform (DWT) and Discrete Kekre transform (DKT). Tecomparison of the hybrid wavelet transforms is also done withthe original orthogonal transforms and their wavelet transforms.The experimentation results have shown that the transformwavelets have given better quality of image compression than therespective original orthogonal transforms but for hybridtransform wavelets the performance is best. Here the hybrid of DCT and DKT gives the best results among the combinations of the four mentioned image transforms used for generating hybridwavelet transforms.

Keywords-Orthogonal transform; Wavelet transform; Hybrid Wavelet transform; Compression.

I.

I

NTRODUCTION

Wavelets are mathematical tools that can be used to extractinformation from many different kinds of data, includingimages [4,5,6,7]. Sets of wavelets are generally needed toanalyze data fully. A set of "complementary" wavelets willreconstruct data without gaps or overlap so that thedeconstruction process is mathematically reversible and is withminimal loss. The wavelets are results of the thought process of many people starting with with Haar's work in the early 20thcentury [19,20]. Generally, wavelets are purposefully crafted tohave specific properties that make them useful for image processing. Wavelets can be combined, using a "shift, multiplyand sum" technique called convolution, with portions of anunknown signal(data) to extract information from the unknownsignal. Wavelet transforms are now being adopted for a vastnumber of applications, often replacing the conventionalFourier transform [23,24,25,26]. They have advantages over traditional fourier methods in analyzing physical situationswhere the signal contains discontinuities and sharpspikes[27,28,29]. In fourier analysis the local properties of thesignal are not detected easily. STFT(Short Time Fourier Transform)[29] was introduced to overcome this difficulty.However it gives local properties at the cost of global properties. Wavelets overcome this shortcoming of Fourier analysis [28,29] as well as STFT. Many areas of physics haveseen this paradigm shift, including molecular dynamics,astrophysics, optics, quantum mechanics etc. This change hasalso occurred in image processing, blood-pressure, heart-rateand ECG analyses, DNA analysis, protein analysis,climatology, general signal processing, speech, facerecognition, computer graphics and multifractal analysis.Wavelet transforms are also starting to be used for communication applications. One use of waveletapproximation is in data compression. Like other transforms,wavelet transforms can be used to transform data then, encodethe transformed data, resulting in effective compression [24].Wavelet compression can be either lossless or lossy. Thewavelet compression methods are adequate for representinghigh-frequency components in two-dimensional images.Earlier wavelets of only Haar transform have been studied.In recent work [4,7,11,13] the wavelets of few orthogonaltransforms alias Walsh [16,17,18], DCT [14,15], Kekre [21,22]and Hartley[1,2,3] are proposed. The wavelet transforms inmany applications are proven to be better than respectiveorthogonal transforms [8,9,10,12]. The paper presents theinnovative hybrid wavelet transform generation method, whichgenerates hybrid wavelet transform of any two orthogonaltransforms. This concept of hybrid wavelet transform canacquire the positive traits from both the orthogonal transformsused to generate it. The hybrid wavelet generation conceptopens up new avenues of selection of orthogonal transforms for hybrid and their use in particular image processing application

80http://sites.google.com/site/ijcsis/ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, 2011

to gain some upper edge over individual orthogonal transformsor respective wavelet transforms. The paper presents the use of hybrid wavelet transforms generated using Discrete WalshTransform (DWT), Discrete Kekre Transform (DKT), DiscreteHartley Transform (DHT) and Discrete Cosine Transform(DCT) for image compression. The experimental results provethat the hybrid wavelet transforms are better than the respectiveorthogonal transforms as well as their wavelet transforms.II.

G

ENERATION OF

H

YBRID

W

AVELET

T

RANSFORM

⎥⎥⎥⎥⎥⎦⎤⎢⎢⎢⎢⎢⎣⎡=

pp p p p p

aaaaaaaaa A

LMMMM
LL

212222111211

⎥⎥⎥⎥⎥⎦⎤⎢⎢⎢⎢⎢⎣⎡=

qqqqqq

bbbbbbbbb B

LMMMM
LL

212222111211

(1)(2)The hybrid wavelet transform matrix of size NxN (say‘T

AB

’) can be generated from two orthogonal transformmatrices ( say A and B respectively with sizes pxp and qxq,where N=p*q=pq) as given by equations 1 and 2.Here first ‘q’number of rows of the hybrid wavelet transform matrix arecalculated as the product of each element of first row of theorthogonal transform A with each of the columns of theorthogonal transform B. For next ‘q’ number of rows of hybridwavelet transform matrix the second row of the orthogonaltransform matrix A is shift rotated after being appended withzeros as shown in equation 2. Similarly the other rows of hybrid wavelet transform matrix are generated (as set of q rowseach time for each of the ‘p-1’ rows of orthogonal transformmatrix A starting from second row upto last row).III.

P

ROPERTIES OF

H

YBRID

W

AVELET

T

RANSFORM

The crossbreed of two orthogonal transforms results intohybrid wavelet transform, which itself satisfies the following properties.

A.

Orthogonal

The transform matrix K is said to be orthogonal if thefollowing condition is satisfied.

[ ] [ ] [ ]

DtK K

=∗

(3)where, D is the diagonal matrix. The hybrid wavelettransform of size NxN generated from any two orthogonaltransforms satisfies this property and hence it is orthogonal.

B.

Non Involutional

An involutionary function is a function that is it’s owninverse. So involutional transform is a transform which isinverse transform of itself. The Hybrid wavelet transform isnon involutional transform

C.

Transform on Vector

The hybrid wavelet transform (say ‘K’) of one-dimensionalvector

q

is given by.

[ ]

qK Q

∗=

(4)And inverse is given by

[ ]

Tj

μ

Ti

μ

ijQtK q

∗∗=

(5)Where Q

ij

is the value at i

th

row, j

th

column of matrix Q andthe term

μ

in normalization factor can be computed as given below through equations 6, 7 and 8.

tABTABTT

μ

=

(6)Such thatB1

μ

A1

μ

T1

μ

=

,

B2

μ

A1

μ

T2

μ

=

,

B3

μ

A1

μ

T3

μ

=

M

Bq

μ

A1

μ

Tq

μ

=

A2

μ

T2q

μ

2Tq

μ

1Tq

μ

===+=+

L

A3

μ

T3q

μ

2T2q

μ

1T2q

μ

===+=+

L

M

A3

μ

T3q

μ

2T2q

μ

11)qT(p

μ

===+=+−

L

(7)Where with reference to equation 1,

A

μ

and

B

μ

can begiven as equation 8.

⎥⎥⎥⎦⎤⎢⎢⎢⎣⎡

==

Ap

μ

000A2

μ

000A1

μ

A

μ

tAA

LMMMM
LL

(8)

81http://sites.google.com/site/ijcsis/ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 6, 2011

⎥⎥⎥⎦⎤⎢⎢⎢⎣⎡

==

Bq

μ

000B2

μ

000B1

μ

B

μ

tBB

LMMMM
LL

D.

Transform on Two-Dimensional Image

The hybrid wavelet transform of two-dimensional image I

isgiven by.

[ ] [ ]

t

K IK Q

∗∗=

(9)And inverse is given by

[ ] [ ]

K Tj

μ

Ti

μ

ijItK q

∗∗∗=

⎟ ⎠ ⎞⎜⎝ ⎛

(10)Where I

ij

is the pixel intensity value at i

th

row, j

th

column of image I and the calculation of the term

μ

in normalizationfactor is as given above through equations 6, 7 and 8.IV.

R

ESULTS AND

D

ISCUSSION

The test bed used in experimentation for proving the worthof hybrid wavelet transform consists of 11 color images of size256x256x3and is shown in figure 1. On each image all thethree alias orthogonal transform, wavelet transform and hybridwavelet transform are applied. In transform domain the highfrequency data is removed and the images are transformedinversely back to spatial domain. To judge the performance of the orthogonal transform, wavelet transform and hybridwavelet transform in compression; the original images arecompared with these modified images (having the data loss ascompression) using mean squared error (MSE). In all size datacompression percentages are considered as 95%, 90%, 85%,80%, 75% and 70%. The average of such MSEs for all imagesfor respective transform and considered percentage of datacompression is taken for performance analysis.

Figure 1: The test bed of eleven original color images belonging to different categories and namely (from left to right and top to bottom) Aishwarya, Balls, Bird,Boat, Flower, Dagdusheth-Ganesh, TajMahal, Strawberry, Scenery, Tiger and Viharlake-Powai.

82http://sites.google.com/site/ijcsis/ISSN 1947-5500