Date post: | 25-Sep-2015 |
Category: |
Documents |
Upload: | hariharankalyan |
View: | 212 times |
Download: | 0 times |
DESIGN OF PIPELINED BASED DCT ARCHITECTURE FOR IMAGE PROCESSING APPLICATIONS
DESIGN OF PIPELINED BASED DCT ARCHITECTURE FOR IMAGE PROCESSING APPLICATIONS
Guided byBy
Dr.R.ManikandanAbhinaya Rajasekaran
SAP115040001
CONTENTS
Abstract
Introduction
Literature survey
Proposed Method
Experimental results
1) Simulation results
2) Synthesis report
Conclusion
ABSTRACT
Discrete Cosine Transform is considered to be an essential tool in image and video compression systems.
The proposed 8 8 transformation matrix contains only zeros and ones which requires only adders, thus avoiding the need for multiplication and shift operations.
This proposed transform requires only 12 additions, which highly reduces the computational complexity and achieves a better image compression when compared to that of the existing approximated DCT.
Another important aspect of the proposed transform is that it provides an efficient area and power optimization.
The model is synthesized in Xilinx Spartan-3 and has been found to have reduction in area and improvement in peak signal to noise ratio when compared with the existing systems.
INTRODUCTION
Discrete Cosine Transform (DCT) has been an active research area over the last decade.
It is used in various image processing applications for reduced bandwidth image and video transmission.
The necessity of using DCT transform is to remove the inter-pixel redundancy from the original image representation.
Reduction of inter-pixel redundancy transforms the image data to a new representation where the average values of transformed data are smaller when compared to the original form.
To achieve higher compression ratio the correlation among the image pixels should be maintained high.
The following properties should be present in an image transform.
Should perform inverse transformation.
Should be capable of de-correlating the original image data.
Should be able to clearly separate the frequency
.
In an image the low frequency contents contain very high visual information when compared to its high frequency counterpart.
The high frequency contents denotes very fine details of an image which is not useful in most of the applications as they are not visible to human eyes.
Therefore to achieve compression these higher frequency contents can be excluded in the coding stage.
Standard DCT
The Discrete Cosine Transform is a Fourier related transform that converts spatial domain to frequency domain and quantize high frequencies more coarsely. The DCT is depicted in the following equation [1],
-------- (1)
Where,
--------- (2)
The equation in (1) can be represented in matrix form as follows:
= ------(3)
Where ci = cos
LITERATURE SURVEY
PAPEROBSERVATIONDRAWBACKSA New Square Wave Transform Based on the DCT[3]DCT transform employs signum function and uses 24 additions for the computationOnly forward DCT can be obtained. Inverse DCT cannot be achieved.A DCT approximation for image compression[4]DCT transform is computed using 22 additions with the help of scaling factorHigh complexity, inefficient power consumption and low speedA low-complexity parametrictransform for image compression[5]DCT transform is performed using 18 additions and 2 bit shit operators by appropriately replacing the coefficients in the signed DCT matrix by 1/2 and 0Requires large number of iterations to compute the DCT valuesPROPOSED METHOD
256X256 Input image is taken
Segmented into 64 8x8 blocks of each
Discrete cosine transform is applied to each 8X8 block
The DCT output is compressed using quantization
Reconstruction of the image is done using de-compression process that involves de-quantization followed by Inverse Discrete Cosine Transform
BLOCK DIAGRAM OF PROPOSED METHOD
8X8 DCT TRANSFORM MATRIX
To design a pipelined DCT architecture where each DCT module can be computed using only 12 additions without any requirement of multipliers.
The path between the two DCT modules is pipelined using Transpose buffer
The first instantiation of the DCT block furnishes a row-wise transform computation of the input image.
The second implementation furnishes a column-wise transformation of the intermediate result
1-D DCT MODULE
In the existing methods the SDCT matrix is modified and reduced from 24 additions to 16 and 14 additions.
This is done by appropriately replacing the matrix with 0s and 1s to improve the PSNR ratio.
In the proposed method similar technique is used to reduce the number of additions to 12.
The following matrix depicts the 12 addition matrix
T = ----------- (5)
The diagonal matrix for the above matrix T is given D=diag which can be merged with the quantization matrix
The following signal flow diagram is built from the matrix (5)
X0
X1
X2
X3
X4
X5
X6
X7
X7
X6
X5
X4
X3
X2
X1
X0
-1
-1
-1
-1
Transposition buffer module
The row wise 8x8 block matrix is computed using the first 1D DCT block which is transposed by the transpose buffer so that that second DCT module can furnish column wise computation.
In the transposition buffer each multiplexer is designed in such a way that it selects the respective row elements assigned to it there by creating the column
The 64 coefficients of the 8x8 matrix is fed into the transpose buffer of which the first element of each row is selected using the multiplexers and is fed as input to the next DCT module
In this way all the elements of each row are transposed into columns.
Transpose buffer module
MUX
D
D
D
D
D
D
D
D
Xj,0
Xj,1
Xj,2
Xj,3
Xj,4
Xj,5
Xj,6
Xj,7
X0,k
Xj,0
Xj,1
Xj,2
Xj,3
Xj,4
Xj,5
Xj,6
Xj,7
Quantization
Quantization process is used to bring variable compression in the image compression process.
The output of the 2D DCT block is fed in to the quantization block where the diagonal matrix D= diag (1, 1, 1, 1, 1, 1, 1, 1) * is merged with quantization matrix Q
Each coefficient of the 8x8 transform matrix coefficients are divided by the coefficient of the quantization matrix and is rounded off to a certain value.
By doing this process, transformed coefficients are made smaller to bring the compression
EXPERIMENTAL RESULTS
DCT TRANSFORMATION USING 14 ADDITIONS
DCT TRANSFORMATION USING 14 ADDITIONS
PROPOSED DCT TRANSFORMATION USING 12 ADDITIONS
IDCT TRANSFORMATION USING 14 ADDITIONS
IDCT TRANSFORMATION USING 14 ADDITIONS
PROPOSED IDCT TRANSFORMATION USING 12 ADDITIONS
SYNTHESIS REPORT
AREA REPORT14-DCT[7]14-DCT[8]Proposed methodNo: of slices580564541No: of slice flip flops897864802No: of 4-input LUTs533469410COMPARISON OF PSNR RATIO OF VARIOUS SYSTEM
(a) Original image, (b)reconstructed using 14 additions[8], (c) reconstructed using 14 additions[7], (d)reconstructed using proposed method
PSNR RATIO OF VARIOUS SYSTEM
(a) DCT using 14 additions[8](b) DCT using 14 additions[7](c) Proposed methodLENA 26.91825.22427.568CAMERA MAN26.12124.92826.736VEGETABLES27.34226.06228.742CONCLUSION AND FUTURE WORK
In this project a new 8 8 transformation matrix, is proposed which requires only 12 additions, thus avoiding the need for multiplication and bit shift operations.
The proposed approximation DCT for image compression is a simple, efficient architecture having lower computational complexity with improvement in the peak signal-to-noise ratio.
The 2D DCT architecture is pipelined to achieve a high throughput while employing a transpose buffer.
The performance of the proposed 2D DCT architecture is calculated using Xilinx tool and the synthesis report was compared with the existing methods.
Based on table 6.1 , 6.2 and 6.3 it can be concluded that the proposed 2D DCT architecture outperforms all the existing architectures in terms of both complexity and speed.
According to the results, the proposed transform has a comparable or better image compression performance than the existing systems.
It has been found to have reduction in area and improvement in peak signal to noise ratio when compared with the existing systems.
This project paves a new research dimension in image compression technique using DCT. This work can be extended by reducing the total number of adders and by implenting it in fpga kit.
PAPER PUBLICATIONS
1.DESIGN OF PIPELINED BASED DCT ARCHITECTURE FOR IMAGE PROCESSING APPLICATIONS
ABHINAYA RAJASEKARAN, R.MANIKANDAN
INTERNATIONAL JOURNAL OF APPLIED ENGINEERING RESEARCH
SCOPUS INDEXED JOURNAL
ACCEPTED
REFERENCES
[1]Ahmed, N., Natarajan, T., & Rao, K. R. Discrete cosine transform. IEEE transactions on Computers, (1974)
[2]W. Chen, C. Smith, and S.C. Fralick, A fast computational algorithm for the discrete cosine transform IEEE Trans. On communication, (1977)
[3]TI Haweel, A new square wave transform based on the DCT. Signal Process (2001).
[4]RJ Cintra, FM Bayer, A DCT approximation for image compression. IEEE Signal Proc Let 18(10), (2011).
[5]S Bouguezel, MO Ahmad, MNS Swamy, A novel transform for image compression, in The 53rd IEEE Int. Midwest Symp. Circuits and Systems (2010)
[6]D Vaithiyanathan, R Seshasayanan, Low power DCT architecture for image compression, in Proceeding of the International Conference on Advanced Computing and Communication Systems (2013)
[7]FM Bayer, RJ Cintra, DCT-like transform for image compression requires 14 additions only. Electron Lett 48(15), (2012).
[8]D Vaithiyanathan, R Seshasayanan, S Anith, K Kunaraj, A low-complexity DCT approximation for image compression with 14 additions only, in Proceeding of the International Conference on Green Computing, Communication and Conservation of Energy (2013)
[9]Vijay Kumar Sharma, Efficient VLSI Architectures for Image Compression Algorithms, NIT thesis, (2012).
[10]Uma Sadhvi Potluri, Arjuna Madanayake, , Renato J. Cintra, Fbio M. Bayer, Sunera Kulasekera and Amila Edirisuriya Improved 8-Point Approximate DCT for Image and Video Compression Requiring Only 14 Additions IEEE Transactions On Circuits And Systems,(2014)
[11]KA Wahid, M Martuza, M Das, C McCrosky, Efficient hardware implementation of 8x8 integer cosine transforms for multiple video codecs. J Real-Time Image Proc, (2013).
Thank you
1D DCT
Using 12
additions
1D DCT
Using 12
additions
T
r
a
n
s
p
o
s
e
b
u
f
f
e
r
x
j,7
x
j,0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
X
j,0
X
j,7
X
0,k
X
7,k
X
0,k
(2D)
X
7,k
(2D)
.
.
.
.
.
.
1D DCT Using 12 additions
1D DCT Using 12 additions
Transpose buffer
xj,0
xj,7
Xj,0
. . . . . .
. . . . . .
. . . . . .
Xj,7
X0,k
X7,k
X0,k(2D)
X7,k(2D)
. . . . . .
1611101624405161
1212141926586055
1413162440576956
1417222951878062
182237566810910377
243555648110411392
49647887103121120101
7292959811210010399
Q
=