# Welcome back

## Find a book, put up your feet, stay awhile

Sign in with Facebook

Sorry, we are unable to log you in via Facebook at this time. Please try again later.

or

Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more

Download

Standard view

Full view

of .

Look up keyword

Like this

Share on social networks

2Activity

×

0 of .

Results for: No results containing your search query

P. 1

A Fast Fractal Image Encoding Based On Haar Wavelet Transform Ratings: (0)|Views: 112|Likes: 1

Published by ijcsis

In order to improve the fractal image encoding, we propose a fast method based on the Haar wavelet transform. This proposed method speed up the fractal image encoding by reducing the size of the domain pool. This reduction uses the Haar wavelet coefficients. The experimental results on the test images show that the proposed method reaches a high speedup factor without decreasing the image quality.

In order to improve the fractal image encoding, we propose a fast method based on the Haar wavelet transform. This proposed method speed up the fractal image encoding by reducing the size of the domain pool. This reduction uses the Haar wavelet coefficients. The experimental results on the test images show that the proposed method reaches a high speedup factor without decreasing the image quality.

See more

See less

https://www.scribd.com/doc/44643187/A-Fast-Fractal-Image-Encoding-Based-On-Haar-Wavelet-Transform

07/30/2011

text

original

A fast fractal image encoding based on Haar wavelettransform

Sofia DoudaDépartement de Mathématiques etInformatique & ENIC, Faculté desSciences et Techniques, UniversitéHassan 1

er

, Settat, Morocco.sofia_douda@yahoo.frAbdallah BagriENIC, Faculté des Sciences etTechniques, Université Hassan 1

er

,Settat, Morocco.Abdelhakim El ImraniLCS, Faculté des Sciences,Université Mohammed V, Rabat,Morocco.

Abstract

—In order to improve the fractal image encoding, wepropose a fast method based on the Haar wavelet transform. Thisproposed method speed up the fractal image encoding byreducing the size of the domain pool. This reduction uses theHaar wavelet coefficients. The experimental results on the testimages show that the proposed method reaches a high speedupfactor without decreasing the image quality.

Keywords- Fractal image compression, PIFS, Haar wavelet transform, SSIM index.

I.

I

NTRODUCTION

Fractal image compression (FIC) is one of the recentmethods of image compression firstly presented by Barnsleyand Jacquin [1-5]. This method is characterized by its highcompression ratio which is achieved with an acceptable imagequality [6], a fast decoding and a multi-resolution property. It isbased on the theory of Iterated Function System (IFS) and onthe collage theorem. Jacquin [3-5] developed the first algorithmof FIC by Local or Partitioned Iterated Function Systems(PIFS) which makes use of local self-similarity propriety inreal images. In FIC, the image is represented through acontractive transformation defined by PIFS for which thedecoded image is approximately its fixed point and close to aninput image.In Jacquin’s algorithm, an input image is partitioned intonon-overlapping sub-blocks R

i

called range blocks, the unionof which covers the whole image. Each range block R

i

is put incorresponding transformation with another part of a differentscale, called domain block, looked for in the image. Thedomain blocks can be obtained by sliding a window of thesame size around the input image to construct the domain pool.The classical encoding method, i.e. full search, is timeconsuming because for every range block the correspondingblock is looked for among all the domain blocks. Severalmethods are proposed to reduce the time encoding. The mostcommon approach is the classification scheme [6-10]. In thisscheme, the domain and the range blocks are grouped in anumber of classes according to their common characteristics.For each range block, comparison is made only for the domainblocks falling into its class. Fisher’s classification method [6]constructed 72 classes for image blocks according to thevariance and intensity. In Wang et al. [10], four types of rangeblocks were defined based on the edge of the image. Jacobs etal. uses skipping adjacent domain blocks [11] and Monro andDudbridge localizes the domain pool relative to a given rangeblock based on the assumption that domain blocks close to thisrange block are well suited to match the given range block [12].Methods based on reduction of the domain pool are alsodeveloped. Saupe’s Lean Domain Pool method discards afraction of domain blocks with the smallest variance [13] andin Hassaballah et al., the domain blocks with high entropies areremoved from the domain pool [14]. Other approaches focusedon improvements of the FIC by tree structure search methods[15, 16], parallel search methods [17, 18] or using two domainpools in two steps of FIC [19]. The spatial correlation in boththe domain pool and the range pool was added to improve theFIC as developed by Truong et al. [20]. Tong [21] proposes anadaptive search algorithm based on the standard deviation(STD). Other approaches based on genetic algorithms are alsoapplied to speed up the FIC [22-23]. In these methods, higherspeedup factor are often associated with some loss of reconstructed image quality. In the present work, a new methodis proposed to reduce the encoding time of FIC using the Haarwavelet transform. It speeds up the time encoding bydiscarding the smooth domain blocks from the domain pool.The type of these blocks is defined using the Haar wavelettransform. A high speedup factor is reached and the imagequality is still preserved.II.

T

HE PROPOSED METHOD BASED ON

H

AAR WAVELETTRANSFORM

A.

The Haar wavelet transform

The Haar Wavelet Transform (HWT) [24] is one of thesimplest and basic transformations from the space domain to alocal frequency domain and it is a very useful tool for signalanalysis and image processing. The HWT decompose a signalinto different components in the frequency domain. One-dimensional HWT decomposes an input sequence into twocomponents (the average component and the detail component)by applying a low-pass filter and a high-pass filter. In the HWTof 2D image of size NxN, a pair of low-pass and high-passfilters is applied separately along the horizontal and verticaldirection to divide the image into four sub-bands of sizeN/2xN/2 (Fig. 1). After one level of decomposition, the low-

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 8, November 201030http://sites.google.com/site/ijcsis/ISSN 1947-5500

low-pass sub-band LL is the multiresolution approximation of the original image, and the other three are high frequency sub-bands representing horizontal, vertical and diagonals edges,respectively. The LL band is again subject to the sameprocedure.LL HLLH HH

Figure 1. The result of 2D image HWT decomposition.

This wavelet decomposition can be repeatedly applied onthe low-low-pass sub-band at a coarser scale unless it only hasone component as shown in fig. 2.Let D be a given image block of size NxN. D can bedecomposed into one component low-pass signal by a logN/log 2 pyramidal HWT. In the case of N=8, D can betransformed to one component by 3 decomposition (Fig. 2).The LL

3

band in level 3 is the multiresolution approximation of LL

2

bands in level 2. The coefficients HL

3,

LH

3

and HH

3

of thehighest level denote the coarsest edges along horizontal,vertical and diagonal directions respectively in level 2.LL

3

HL

3

LH

3

HH

3

HL

2

LH

2

HH

2

HL

1

LH

1

HH

1

Figure 2. The result of three level HWT pyramidal decomposition of animage block of size 8x8.

We will refer to these coefficients obtained at the highestlevel hereafter as WH

D

for HL

3

, WV

D

for LH

3

and HH

3

forWD

D

.If both WH

D

and WV

D

are small, then the block D tends tohave less edge structure (smooth block). When a block has highdegree of edge structure, either WH

D

or WV

D

will be large. If WH

D

is larger, D will have horizontal edge properties. On theother hand, if WV

D

is larger, then D will have vertical edgeproperties. Finally, those blocks with high magnitudes of WH

D

and/or WV

D

are designed as heterogeneous domain blocks. Thetype of each block D is determined as follows:

WW

WH

WVDD

ifTandTthenDisasmoothdomainblock elseDisaheterogeneous domainblock

< <

(1)where |.| denotes the absolute value of its variable and T

W

is athreshold.Thus, an image block D can be determined as belonging tosmooth or heterogeneous type by using its vertical coefficientWV

D

and its horizontal coefficient WH

D

obtained by apyramidal HWT at the highest level.The computation of WH

D

and WV

D

do not require thecalculation of other wavelets coefficients. Indeed, let D be animage block of size 4x4 represented as follows:1 2 3 45 6 7 89 10 11 1213 14 15 16Then, analyzing the expression of the coefficients WH

D

andWV

D

obtained after two pyramidal HWT, allow us to find thefollowing simplified formula:

D

ABWH16

−=

(2)

D

CDWV16

−=

(3)where

A12563478

= + + + + + + +

,

B910131411121516

= + + + + + + +

,

C1256101314

= + + + + + +

,

D347811121516

= + + + + + + +

.

B.

The proposed method

The proposed method is aimed to reduce the encoding timeby reducing the cardinal of the domain pool. As only a fractionof the domain pool is used in fractal encoding and the set of theused blocks is localized along edges and in the regions of highcontrast of the image (designed as heterogeneous blocks), it’spossible to reduce the cardinal of the domain pool bydiscarding the smooth domain blocks. Therefore, each rangebloc is compared only to the heterogeneous domain blocks.This method of reduction of the domain pool is simple sinceonly few computations are required to calculate the coefficientsWH

D

and WV

D

of a domain block D to classify it as smooth orheterogeneous.The threshold T

W

can be fixed or chosen in an adaptiveway. Determining T

W

adaptively allow us to choose thespeedup ratio. The main idea is to set the thresholds such that afraction

α

of the domain pool can be discarded. The value of

α

can be one third, tow thirds,... of the domain pool. Due to thefact that the encoding time depends on the number of comparisons between range and domain blocks, the speedupratio can be estimated.The determination of the threshold T

W

, which depends onthe fraction

α

of the domain pool to be eliminated, issummarised as follows:

•

For each domain block D, calculate the Haar waveletcoefficient WH

D

and WV

D

.

Set S

D

=max(|WH

D

|,|WV

D

|).

•

Sort all the values of S

D

in increasing order.

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 8, November 201031http://sites.google.com/site/ijcsis/ISSN 1947-5500

•

Find S* corresponding to the value of

α

. Set thethreshold T

W

=S*.Due to the fact that we apply our method in the case of aquadtree partitioning, we choose different thresholds for everysize of the domain blocks.The first steps of the proposed method are as follows:

•

Choose a value of

α

.

•

Construct the domain pool.

•

Compute the Haar wavelet coefficients |WH

D

| and|WV

D

| for each domain block D.

•

Determine the threshold T

W

for each size domainblock.

•

Remove from the domain pool the smooth domainblocks.III.

E

XPERIMENTAL RESULTS

The different tests are performed on three 256x256 images,represented in fig. 3, with 8 bpp on PC with Intel Pentium Dual2.16 Ghz processor and 2 GO of RAM. The quadtreepartitioning [6] is adopted for the FIC. The encoding time ismeasured in seconds. The rate of compression is represented bythe compression ratio (CR), i.e. the size of the original imagedivided by the size of the compressed image. The speedupfactor (SF) of a particular method can be defined as the ratio of the time taken in full search to that of the said method, i.e.,

Time taken in full searchSFTime taken in a particular method

=

(4)The image quality is measured by the peak signal to noiseratio (PSNR) and the structural similarity Measure (SSIM)index [25].The PSNR of two images X and Y of sizes N, is defined asfollows:

2

255PSNR10xlogMSE

=

(5)where

N2iii1

N

1MSE(xy)

=

= −

∑

(6)x

i

and y

i

are the gray levels of pixel of the original imageand the distorted image respectively.The SSIM index is a method for measuring the similaritybetween two images x and y defined by Wang [25] as follows:

xyxy212222xy1xy2

(2C)(2C)SSIM(x,y)(C)(C)

+

µ µ + σ +=µ +µ + σ σ +

(7)where

xi

1xN

µ =

∑

,

yi

1yN

µ =

∑

,

122

1xixN1

((x))

−

σ = −µ

∑

,

122

1yiyN1

((y))

−

σ = −µ

∑

,

1xyixiyN1

(x)(y))

−

σ = −µ −µ

∑

.C1 and C2 are positive constants chosen to prevent unstablemeasurement when

22xy

()

µ +µ

or

22)xy

(

+

σ σ

is close to zero. Theyare defined in [25] as:C

1

= (K

1

L)

2

, C2= (K

2

L)

2

(8)where L is the dynamic range of pixel values (L= 255 for 8-bitgray scale images). K

1

and K

2

are the same as in [20]: K1=0.01 and K2= 0.03.In the present work, we use a mean SSIM (MSSIM) indexto evaluate the overall image quality:

Miii1

1MSSIM(X,Y)SSIM(x,y)M

=

=

∑

(9)where X and Y are the original and the distorted imagesrespectively; x

i

and y

i

are the image contents at the i

th

localwindow of size 8x8 and M is the number of local windows of the image.

(a)(b)

(IJCSIS) International Journal of Computer Science and Information Security,Vol. 8, No. 8, November 201032http://sites.google.com/site/ijcsis/ISSN 1947-5500

- Read and print without ads
- Download to keep your version
- Edit, email or read offline

© Copyright 2015 Scribd Inc.

Language

Choose the language in which you want to experience Scribd:

Sign in with Facebook

Sorry, we are unable to log you in via Facebook at this time. Please try again later.

or

Password Reset Email Sent

Join with Facebook

Sorry, we are unable to log you in via Facebook at this time. Please try again later.

or

By joining, you agree to our

read free for two weeks

Personalized recommendationsbased on books you love

Syncing across all your devices

Join with Facebook

or Join with EmailSorry, we are unable to log you in via Facebook at this time. Please try again later.

Already a member? Sign in.

By joining, you agree to our

to download

Personalized recommendationsbased on books you love

Syncing across all your devices

Continue with Facebook

Sign inJoin with emailSorry, we are unable to log you in via Facebook at this time. Please try again later.

By joining, you agree to our

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

CANCEL

OK

You've been reading!

NO, THANKS

OK

scribd