International Journal on Future Revolution in Computer Science & Communication Engineering ISSN: 2454-4248

Volume: 3 Issue: 10 155 – 159

_______________________________________________________________________________________________
A Novel Hybrid Linear Predictive Coding – Discrete Cosine Transform based Compression

Sandip Mehta
Department of Electrical and Electronics Engineering
JIET Group of Institutions
Jodhpur, Rajasthan, India
sandip.mehta@jietjodhpur.ac.in

Abstract—Image compression is a type of data compression applied to digital images, for reducing the cost for their storage and transmission.
Algorithms may take advantage of Algorithms may take advantage of visual perception and the statistical properties of image data to provide
superior results compared with generic compression methods. Image compression may be lossy or lossless. Lossless compression is preferred for
archival purposes and often for medical imaging, technical drawings, clip art, or comics. Lossy compression methods, especially when used at
low bit rates, introduce compression artifacts. Lossy methods are especially suitable for natural images such as photographs in applications
where minor (sometimes imperceptible) loss of fidelity is acceptable to achieve a substantial reduction in bit rate. Lossy compression that
produces negligible differences may be called visually lossless. On useful technique used for lossy compression is the discrete cosine transform
(DCT) that helps separate the image into parts (or spectral sub-bands) of differing importance (with respect to the image's visual quality). The
DCT is similar to the discrete Fourier transform in the sense that it transforms a signal or image from the spatial domain to the frequency
domain. This paper proposes a hybrid lossy compression technique using Linear Predictive Coding (LPC) and Discrete Cosine Transform (DCT)
to provide superior compression ratios.
Keywords- Lossless Predictive Coding , Compression, Discrete Cosine Transform, Inverse Discrete Cosine Transform, PSNR
__________________________________________________*****_________________________________________________

the reconstruction is traded with efficiency of compression.

I. INTRODUCTION Lossy compression is mainly used to compress multimedia
Digital Image processing is a technique to perform some applications such as image, audio and video which are used for
operations on an image so as to get an enhanced image or to activity of internet and media. The most important work is
extract some useful information from it. It is a form of signal reducing redundancy and irrelevancy.
processing in which input is an image and output may be
There are many lossy techniques that operate directly on the
image or characteristics/features associated with that image.
pixels of an image and thus are spatial domain methods. There
Nowadays, image processing is among rapidly growing
is, however, a family of popular compression standards that
technologies. It forms core research area within engineering
are based on transforms. Transform coding is a type of data
and computer science disciplines too. Image processing
compression for natural data like audio signals or
basically includes the following three steps viz. importing the
photographic images. The transform is typically lossy,
image via image acquisition tools, analysing and manipulating
resulting in a lower quality copy of the original input. In
the image, and obtaining the output in which the result can be
transform coding, knowledge of the application is used to
altered image or report that is based on image analysis.
choose information to discard, thereby lowering its bandwidth.
There are two types of methods used for image processing The remaining information can then be compressed via a
namely, analogue and digital image processing. Analogue variety of methods, when the output is decoded, the result may
image processing can be used for the hard copies like printouts not be identical to the original input, but is expected to be
and photographs. Image analysts use various fundamentals of close enough for the purpose of the application. Several
interpretation while using these visual techniques. Digital transform coding algorithms are found in the literature for
image processing techniques help in manipulation of the obtaining a good level of compression including Discrete
digital images by using computers. The three general phases Cosine transforms [21],[22],[28]-[31], Karhunen-Loeve
that all types of data have to undergo while using digital Transforms [23] and Wavelet Transforms [24]-[27],[32]. This
technique are pre-processing, enhancement, and display, paper proposes a transform based coding method using the
information extraction. Discrete Cosine Transform and Lossless Predictive Coding, an
improvement over the standard JPEG algorithm.
The objective of digital image compression [1]-[10] is to
minimize the redundancy of an image and to store or transmit The rest of the paper is arranged as follows. The methodology
data through the network in an efficient form. It is also used to has been given in section II while the experimental results
reduce the size in bytes without affecting the quality of the have been presented in section III. The conclusion is given in
original image. It has been classified into lossless [11]-[18] section IV.
and lossy image compression [19]-[31]. In lossless image
II. METHODOLOGY
compression, original image is exactly reconstructed after the
decompression process. It is mainly preferred for archival The Discrete Cosine Transform and the Lossless Predictive
purpose and often for medical images, icons, clip arts, comics Coding are hybridized in a common framework to produce a
or technical drawings. In lossy compression, the constructed new lossy compression method. The block diagram of the
image is not exactly the same as the original image. Rather, proposed methodology is shown in Fig. 1.

155
IJFRCSCE | October 2017, Available @ http://www.ijfrcsce.org
_______________________________________________________________________________________
International Journal on Future Revolution in Computer Science & Communication Engineering ISSN: 2454-4248
Volume: 3 Issue: 10 155 – 159
_______________________________________________________________________________________________

Modified 8 8 Normalizer/ Symbol

Input DCT Compressed
LPC block quantizer encoder image
image
extractor

(a)

Compressed Symbol Inverse 8 8 Inverse Reconstructed

Denormalizer DCT block modified
image decoder image
merger LPC

(b)
Fig. 1. Block Diagram of Hybrid LPC-DCT Compression Method (a) Encoder (b) Decoder
Stage 3: Level shifting of the image for better coding: As each
As shown in Fig. 1, the proposed compression method has
8x8 block or subimage is processed, its 64 pixels are level
been organized in two parts: the coding part and the decoding shifted by subtracting 2N-1 where 2Nis the number of gray
part. The steps involved in the HLPCDCT method for both levels in the image. Level shifting is done in order to make the
coding and decoding are as follows: data fit the discrete cosine transform. This results in the 8-bit
pixels falling in the range of -127 to 128 thereby making the
A. CODING data symmetric across zero. This is useful for DCT as any
Stage 1: Application of the modified Lossless Prediction symmetry that is exposed will lead towards better entropy
coding. In the normal method, the first column was taken as compression. Effectively, this shifts the DC coefficients to fall
reference and the differences were along the horizontal more in line with value of the AC coefficients. It should
direction. In the modified case, after processing in the however be noted that the AC coefficients produced by the
horizontal direction, the first column is itself predictive coded DCT are not affected in any way by this his level shifting.
which further improves the performance of the system.
Stage 4: Computation of the 2-D DCT of the level-shifted
Stage 2: Sub-division of the input image into non-overlapping image: The 2D DCT of the input image X(i j) is taken to get
pixel blocks of size of 8x8, which are then subsequently G(k, l). The DCT matrix of order 8 is given as follows:
processed left to right, top to bottom. The reason for selecting
8x8 block size can be understood from Fig.6. 4 which 0.3536 0.3536 0.3536 0.3536 0.3536 0.3536 0.3536
0.3536
illustrates graphically the impact of subimage size on DCT
coding reconstruction error. The data plotted are obtained by 0.4904 0.2778 0.0975 -0.0975 -0.2778 -0.4157 -0.4904
0.4157
dividing the 512x512, 8-bit monochrome image ‗Lena‘ into
subimages of size n x n, for n= 2,4,8, 16,…, 256,512, 0.4619 0.1913 -0.1913 -0.4619 -0.4619 -0.1913 0.1913 0.4619
computing the transform of each subimage, truncating 75% of
the resulting coefficients, and taking the inverse transform of 0.4157 -0.0975 -0.4904 -0.2778 0.2778 0.4904 0.0975 -0.4157
the truncated arrays. It can be seen that the cosine curve
flattens as the size of the subimage becomes greater than 8 x 8. 0.3536 -0.3536 -0.3536 0.3536 0.3536 -0.3536 -0.3536 0.3536
Therefore, the optimum size of the subimage is 8 x 8.
0.2778 -0.4904 0.0975 0.4157 -0.4157 -0.0975 0.4904 -0.2778

0.1913 -0.4619 0.4619 -0.1913 -0.1913 0.4619 -0.4619 0.1913

0.0975 -0.2778 0.4157 -0.4904 0.4904 -0.4157 0.2778 -0.0975

The 2-D DCT of X(i j) is given by

G(k, l) = round (h x X(i j) x hT ) (1)
Stage 5: Normalization and Quantization of DCT coefficients:
The coefficients obtained ins step 4 are then simultaneously
normalized and quantized in accordance with

(2)
Fig. 2. Reconstruction Error versus Subimage Size for DCT
where the quantization matrix q(u,v) is given as

156
IJFRCSCE | October 2017, Available @ http://www.ijfrcsce.org
_______________________________________________________________________________________
International Journal on Future Revolution in Computer Science & Communication Engineering ISSN: 2454-4248
Volume: 3 Issue: 10 155 – 159
_______________________________________________________________________________________________
may be adapted to the characteristics of the image being
16 11 10 16 24 40 51 61 compressed.
12 12 14 19 26 58 60 55 B. DECODING

14 13 16 24 40 57 69 56 Stage 1: To decompress a compressed subimage, the decoder

must first recreate the normalized transform coefficients that
14 17 22 29 51 87 80 62 led to the compressed bit stream. Because a Huffman-coded
binary sequence is instantaneous and uniquely decodable, this
q (u,v) = 18 22 37 56 68 109 103 77 step is easily accomplished in a simple look-up table manner.

24 35 55 64 81 104 113 92 Stage 2: The obtained array is now denormalized by using the
following equation:
49 64 78 87 103 121 12 101
* q(u,v) (3)
72 92 95 98 112 100 103 99 Stage 3: The partial reconstructed subimage is obtained by
taking the inverse DCT of the denormalized array in
Fig. 3. A Typical Normalization Matrix accordance with equation 2 and 3 and then level shifting each
inverse transformed pixel by 2N.
and Q (u, v) for u,v = 0, 1, 2, 3,….7 are the resulting
normalized and quantized coefficients , G(k,l) is the DCT of Stage 4: Finally the inverse LPC operation is applied to obtain
an 8x8 block of image f(x,y), and q(u,v) is a transform the complete reconstructed image.
normalization array. By scaling q(u,v) , a variety of
compression ratios and reconstructed image qualities can be IV. EXPERIMENTAL RESULTS
achieved.
The performance of this technique is measured using
Stage 6: Pattern directed reordering of the normalized compression ratio (CR), Peak-signal-noise-ratio (PSNR) and
coefficients: After each block‘s DCT coefficients are root-mean- square error (RMSE). Visual inspection is also
quantized, the elements of are reordered in carried out on the compressed images as to judge the
accordance with the zigzag pattern of Fig. 4. effectiveness of the compression method. A wide range of
quality factor ranging from 1 to 12 was taken for the six
0 1 5 6 14 15 27 28 images and the results tabulated and compared with the
standard JPEG results and shown in Table 1.
2 4 7 13 16 26 29 42 TABLE 1
Compression Results for ‗Lena‘ Image
3 8 12 17 25 30 41 43
Proposed Proposed
JPEG JPEG
Quality Method Method
9 11 18 24 31 40 44 53
Compression Ratio(CR) PSNR
10 19 23 32 39 45 52 54 1 9.8729 10.5195 41.3594 39.9417

2 18.1716 19.7079 34.7799 33.4624

20 22 33 38 46 51 55 60
3 24.8195 26.1038 32.8510 31.2459
21 34 37 47 50 56 59 61
4 26.9695 29.0249 31.6603 30.6712
35 36 48 49 57 58 62 63 5 32.9741 34.6697 30.7721 29.5641

Fig. 4. Zigzag Pattern 6 38.9979 41.5760 29.9159 28.9548

7 43.0733 44.5973 29.2170 28.2354

Since the resulting one-dimensionally reordered array (of
quantized coefficients) is qualitatively arranged according to 8 46.6614 50.9834 28.6674 27.7851
increasing spatial frequency, the symbol encoder of Fig. 1 is 9 51.8276 54.7701 27.9780 27.9815
designed to take advantage of the long runs of zeros that
normally result from the reordering. In particular, the nonzero 10 56.6186 60.9726 27.4986 26.9842
AC coefficients (i.e. all except u= v = 0) are coded 11 59.9049 64.8624 27.0469 26.2314
using a variable length code that defines the coefficients value
and number of preceding zeros. 12 64.1881 69.0248 26.7658 26.2314

Stage 7: Application of Huffman Coding: The DC coefficients

are difference code relative to the DC coefficient of the It can be seen from the above table that the proposed algorithm
previous subimage. Default AC and DC Huffman coding performs significantly better than the conventional JPEG at
tables are provided by the standard, but custom tables and almost all levels of quality factor. Fig.5 illustrates the
normalization arrays can be or have been constructed, which qualitative results at different quality levels for ‗Lena‘ image.
157
IJFRCSCE | October 2017, Available @ http://www.ijfrcsce.org
_______________________________________________________________________________________
International Journal on Future Revolution in Computer Science & Communication Engineering ISSN: 2454-4248
Volume: 3 Issue: 10 155 – 159
_______________________________________________________________________________________________

REFERENCES
[1] R. C. Gonzalez and R. E. Woods, ―Digital Image Processing‖, 2nd
edition, Prentice Hall, 2002.
[2] R. C. Gonzalez, R. E. Woods and S. L. Eddins, ―Digital Image
Processing Using MATLAB‖, Pearson, 2004.
[3] K. R. Castleman, ―Digitial Image Processing‖, Pearson, 1996.
[4] M. Petrou and C. Petrou, ―Image Processing –The Fundamentals‖, 2nd
edition, Wiley, 2010.
[5] A. Bovik, ―Handbook of Image and Video Processing‖, Academic
Press, 2000.
Quality = 1 Quality =3
[6] O. Marques, ―Practical Image and Video Processing Using MATLAB‖,
(a) (b) IEEE Press, Wiley, 2011.
[7] S. Jayaraman, S. Esakkirajan and T Veerakumar, ―Digital Image
Processing‖, Tata McGraw Hill, 2009.
[8] K. Sayood, ―Introduction to Data Compression‖, Elsevier, 2006.
[9] J. M. Blackledge and M. J. Turner, ―Image Processing II- Mathematical
Methods, Algorithms and Applications‖, Horwood Publishing Series:
Mathematics and Applications, 2000.
[10] J. M. Blackledge and M. J. Turner, ―Image Processing III-
Mathematical Methods, Algorithms and Applications‘, Horwood
Publishing Series: Mathematics and Applications, 2001.
[11] S. W. Golomb, "Run-length encodings", IEEE Transactions on
Information Theory, Vol. IT-12, pp. 399401, July 1966.
Quality = 5 Quality = 7
[12] G. Motta, J. A. Storer, and B. Carpentieri, "Lossless image coding via
(c) (d)
adaptive linear prediction and classification", Proceedings of the IEEE,
Vol. 88, No.11, pp. 1790-1796, 2000.
[13] Aleksej Avramovic and Salvica Savic,"Lossless Predictive
Compression of Medical Images", Serbian Journal of Electrical
Engineering, Vol. 8, no.1, pp. 27-36, Feb. 2011.
[14] I. H. Witten, R. Neal, and J. M. Cleary, "Arithmetic Coding for Data
Compression", Communications of the ACM, Vol. 30, no. 6, pp.
520540, June 1987.
[15] D. A. Huffman, "A method for the construction of minimum
redundancy codes", Proceedings of the IRE, Vol. 40, no. 9, pp.
10981101, September 1952.
[16] R. Hashemian , ―High Speed Search and Memory Efficient Huffman
Quality = 9 Quality = 11 Coding," IEEE Inter. Symp. Circuit Syst. May 3-6, 1993.
(e) (f) [17] R. Hashemian,"Design and Hardware Implementation of a Memory
Fig. 5. Restoration Results of HLPC-DCTC for ‘Lena’ Image for efficient Huffman Decoding", EEE Consumer Elec. August Vol. 40,
Different Values of Quality (a) Quality = 1 (b) Quality =3 No.3, 1994, pp. 345-352.
(c) Quality =5 (d) Quality =7 (e) Quality =9 (f) Quality = 11 [18] S. B. Choi, Moon Ho Lee "A Fast Huffman Decoder via Pattem
Matching", 1994 IEEE Intemational Workshop on ISPACS, 1994, PP
The hybrid Linear Predictive Coding –Discrete Cosine 134-138.
Transform compression technique proposed in this section is [19] P. C. Cosman, E. A. Riskin, R. M. Gray, "Combining vector
competitively better than the JPEG algorithm at all levels of quantization and histogram equalization", Informal. Processing Manag.,
quality. However, like the JPEG algorithm, it also suffers vol. 28, no. 6, pp. 681-686, Nov.—Dec. 1992.
from blocking artifacts as the value of quality increases. [20] P. A. Chou, T. Lookabaugh, R. M. Gray, "Entropy-constrained vector
quantization", IEEE Trans. Acoust. Speech Signal Processing, vol. 37,
pp. 31-42, Jan. 1989.
IV. CONCLUSION [21] A. J. Ahumada, H. A. Peterson, "Luminance-model-based DCT
A Novel Hybrid lossy compression technique has been quantization for color image compression", Human Vision Visual
Processing and Digital Display III, vol. 1666, pp. 365-374, 1992.
proposed in this technique which incorporates the Linear
[22] H. A. Peterson, A. J. Ahumada, A. B. Watson, "An improved detection
Predictive Coding and the Discrete Cosine Transform. The model for DCT coefficient quantization", SPIE Proceeding, vol. 1913,
simulations were carried out for a number of benchmark pp. 191-201, 1993.
images including Lena, Peppers, Barbara, Lighthouse, [23] A. K. Jain, "A fast Karhunen-Loeve transform for a class of random
Mandrill and Grass. In each of these cases, it was found that processes", IEEE Trans. Comm., vol. COM-24, pp. 1023-1029, Sept.
the results obtained from this technique were superior to that 1976.
obtained from JPEG at various levels of quality factor. [24] I. Daubechies, ―Ten Lectures on Wavelets‖, Society for Industrial and
Applied Mathematics, Philadelphia, 1992.
However, like the JPEG algorithm, it also suffers from [25] S. Mallat, ―A Wavelet Tour of Signal Processing‖, 2nd edition,
artefacts. Future work may be directed at removing these Academic Press, 1999.
artifacts and reducing the processing time. [26] S. Mallat, ―A Theory for Multiresolution Signal Decomposition: The
Wavelet Representation,‖ IEEE Trans. Pattern Anal. Mach. Intell.,
Vol. 11, No. 7, pp. 674–693, Jul. 1989.
158
IJFRCSCE | October 2017, Available @ http://www.ijfrcsce.org
_______________________________________________________________________________________
International Journal on Future Revolution in Computer Science & Communication Engineering ISSN: 2454-4248
Volume: 3 Issue: 10 155 – 159
_______________________________________________________________________________________________

[27] R. M. Rao and A. S. Bopardikar, ―Wavelet Transforms- Introduction to

Theory and Applications‖, Pearson Education, 2002.
[28] W.-H. Chen, C. Smith, S. Fralick, "A fast computational algorithm for
the discrete cosine transform", IEEE Trans. Commun., vol. 25, no. 9,
pp. 1004-1009, Sep. 1997.
[29] Z. Xiong, O. G. Guleryuz, M. T. Orchard, "A DCT-based embedded
image coder", IEEE Signal Process. Lett., vol. 3, pp. 289-290, Nov.
1996.
[30] D. Nister, C. Christopoulos, "An embedded DCT-based still image
coding algorithm", IEEE Signal Process. Lett., vol. 5, pp. 135-137, Jun.
1998.
[31] P. Telagarapu, V.J. Naveen, A.L Prasanthi, G.V. Santhi, "Image
compression using DCT and wavelet transformations", Int. J. of Signal
Processing Image Processing and Pattern Recognition, vol. 4, no. 3,
2011.

159
IJFRCSCE | October 2017, Available @ http://www.ijfrcsce.org
_______________________________________________________________________________________