Dictionary based Image Compression via Sparse Representation

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 7, No. 4, August 2017, pp. 1964~1972
ISSN: 2088-8708, DOI: 10.11591/ijece.v7i4.pp1964-1972  1964
Journal homepage: http://iaesjournal.com/online/index.php/IJECE
Dictionary based Image Compression via Sparse Representation
Arabinda Sahoo1
, Pranati Das2
1
Department of ECE, ITER, Siksha „O‟ Anusandhan University, Bhubaneswar, Odisha, India
2
Department of Electrical Engineering, IGIT, Sarang, Odisha, India
Article Info ABSTRACT
Article history:
Received Feb 27, 2017
Revised May 25, 2017
Accepted Jun 15, 2017
Nowadays image compression has become a necessity due to a large volume
of images. For efficient use of storage space and data transmission, it
becomes essential to compress the image. In this paper, we propose a
dictionary based image compression framework via sparse representation,
with the construction of a trained over-complete dictionary. The over-
complete dictionary is trained using the intra-prediction residuals obtained
from different images and is applied for sparse representation. In this
method, the current image block is first predicted from its spatially
neighboring blocks, and then the prediction residuals are encoded via sparse
representation. Sparse approximation algorithm and the trained over-
complete dictionary are applied for sparse representation of prediction
residuals. The detail coefficients obtained from sparse representation are
used for encoding. Experimental result shows that the proposed method
yields both improved coding efficiency and image quality as compared to
some state-of-the-art image compression methods.
Keyword:
Dictionary learning
Image compression
Intra prediction
K-SVD
Sparse representation
Copyright © 2017 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Arabinda Sahoo,
Departement of Electronics and Communication Engineering,
ITER, Siksha „O‟ Anusandhan University,
Bhubaneswar, Odisha, India.
Email: arubabu123@yahoo.co.in
1. INTRODUCTION
Images compression [1] has always been important, and nowadays it becomes essential due to the
huge requirement of image storage and transfer. Over the last two decades, numerous and diverse image
compression methods [2-6] have been proposed. The most widely used methods are based on transform-
based coding. Based on transform-based approaches by far many image compression standards like JPEG [2]
and JPEG2000 [3] have been developed. The widely used JPEG and JPEG2000 standards utilize discrete
cosine transform (DCT) and wavelet transform to compressively represent images. However, JPEG and
JPEG2000 do not take spatial correlation of neighboring blocks into consideration for more compact
representation of images [7]. In recent years, intra-prediction [8], [9] schemes serve as a promising direction,
which use neighboring blocks in the compression process for more compression. The idea of intra-prediction
is to predict the unknown image block based on the knowledge of decoded neighboring blocks. A good
prediction reduces the overall coding rate.
However, these compression methods suffer some limitations, because they are not able to
efficiently compress some specific classes of images. They are not able to sparsely represent complex
characteristics of an image. To overcome these limitations, sparse representations [10] has been evolving in
recent years. Sparse representation is a very effective tool for compressing a large variety of images. This is
possible due to the fact that the images can be sparse or compressible with respect to some basis or dictionary
[11]. Thus, sparse representation provides a potential for effective image compression. An image is
compressible or not, it depends on the dictionary, so the design of dictionary is vital in sparse representation.

IJECE ISSN: 2088-8708 
Dictionary based Image Compression via Sparse Representation (Arabinda Sahoo)
1965
In contrast to fixed DCT and wavelet dictionary, the latest trend of image compression techniques is
extended to design trained dictionaries [12]. Numerous dictionary based image compression methods have
been proposed for sparse representation. In several recently published works, the use of learned over-
complete dictionaries in image compression has shown promising results at low bit rate as compared to fixed
dictionaries. The initial proposal towards dictionary based image compression was proposed by Bryt and
Elad [13]. They proposed an algorithm for the compression of facial image based on the learned K-SVD [14]
over-complete dictionary. Though this method outperforms JPEG and JPEG2000 but this method is limited
to compressing facial images. In [15], the author proposed an image compression method based on the
iteration-tuned dictionary (ITD). In this scheme, the dictionary consists of a layer structure with each layer
trained for a specific class of images and carries a separate dictionary matrix. This method is shown to
outperform K-SVD over-complete dictionary method but it is employed to compress a specific class of
images. In [16], the author proposed an adaptive dictionary design method for the fingerprint image
compression. They used a set of fingerprint image to learn a dictionary. In [17], the author proposed a
dictionary based method for compressing surveillance image.
Experimental result shows these above proposed algorithms outperform JPEG and JPEG2000.
However, most of the above mentioned compression scheme is either related to facial images or to some
specific class of images, while there is a lack of research on general arbitrary images. So, the main challenge
of above schemes is the compression of general arbitrary images. To address this, in this paper we proposed a
novel image compression scheme for arbitrary images. In this scheme, an efficient trained over-complete
dictionary is integrated into the intra-prediction framework. The conventional transform-domain
representation of intra-prediction scheme is replaced by a trained over-complete dictionary. We trained a
dictionary offline using the residuals obtained from intra-prediction. This dictionary can sparsely represent
the complex characteristics of the residual block. The coefficients and indices of appropriate dictionary
element obtained from sparse representation are transmitted for encoding. Since the dictionary is shared at
both encoder and decoder, only coefficients and indices of dictionary elements need to be encoded, which
compress the image significantly. Experimental results on the arbitrary images shows that the proposed
method yields both improved coding efficiency and image quality as compared to JPEG and JPEG 2000.
The rest of the paper is organized as follows. In Section 2, we present some preliminaries on sparse
representation and dictionary design. The proposed image compression method is introduced in section 3.
Section 4 illustrates experimental results and discussion. Finally, Section 5 concludes the paper.
2. PRELIMINARIES
In sparse representation [18], a signal can be represented by a linear combination of a small
number of signals known as atoms taken from an over-complete dictionary . It is called sparse
representation as it employs only a few number dictionary atoms or elements to represent the signal. A signal
is said to be compressible if it can be represented by few dictionary atoms. Mathematically sparse
representation can be expressed as:
(1)
The solution vector contains the coefficients of the signal . Where D is one n×k matrix with n < k
called over-complete dictionary and each column of D is called an atom. If and each atom of D are treated
as a signal then can be represented as a linear combination of atoms of D. This linear combination can be
expressed as solution vector . Due to over complete nature of D, an infinite number of solution exist for .
In last decade, various sparse approximation algorithms have been proposed to find out the sparse solution
for . The sparse approximation algorithm always aims to represent in terms of minimum number atoms.
Mathematically, this can be expressed as solving Equation (1) such that the solution contains minimum
number of non-zero elements. A signal is said to be compressible if the number of non-zero elements in is
very less as compared to number of elements of . The sparse representation of may be either exact
or approximate, satisfying .Where p is the norm.Typical norms used in the
approximation are 1, 2 or ∞. Normally p = 2 is taken in image compression. Mathematically, the sprarsest
representation is the solution of either Equations (2) or (3)
(2)
or
(3)

 ISSN: 2088-8708
IJECE Vol. 7, No. 4, August 2017 : 1964 – 1972
1966
where is the l0 norm, represents the number of non-zero elements in solution vector x. Some of the well
known algorithms used to find sparse representation are: Matching pursuit (MP) [19], Orthogonal Matching
Pursuit (OMP) [20] and Complimentary matching pursuit (CMP) [21]. In this paper, we focus on OMP
algorithm due to its efficiency.
The dictionary based image compression can also be effectively modeled by Equations (2) and (3).
In image compression, we consider a set of k image blocks of size n pixels with n < k, ordered
lexicographically as column vectors of dictionary D. A column vector y is obtained from an image block of
size n pixels. In sparse representation, the problem is to find out the solution vector which will represent y
with least number of dictionary elements. Indeed, compression of image patch y can be achieved by
transmission of nonzero elements of vector , by specifying their coefficients and indices.
The dictionary plays an important role in a successful image compression modeling via sparse
representation. Am image is compressible if it can be represented by few number of dictionary elements. The
dictionary can either be chosen as a prespecified set of images or designed by adapting its contents to fit a
given set of images. The objective of dictionary design is to train the dictionary which able to represent a
signal set sparsely [12]. Given an image set { } , dictionary design aims to find the best dictionary D
that gives rise to sparse solution for each . In other words, there exists D, such that solving Equation (2)
for each gives a sparse representation . The minimization problem to find the best dictionary for sparse
representation of Y in the given sparsity constraint T0 can be represented by:
(4)
The dictionary is trained to provide a better representation of the actual signal when the number of
dictionary elements used to represent it is less than or equal to T0.Various algorithms have been developed to
train over complete dictionaries for sparse signal representation. The K-SVD algorithm [14] is very efficient
and it works well with different sparse approximation algorithm. K-SVD algorithm iteratively updates the
dictionary atoms to better fit the data. In this paper, we focus on K-SVD algorithm to train the dictionary and
OMP algorithm for sparse representation.
(a) Encoder
(b) Decoder
Figure 1. Detailed block diagram of the proposed method

IJECE ISSN: 2088-8708 
1967
3. PROPOSED METHOD
The proposed image compression method consists of four main processes: intra prediction,
dictionary training, sparse representation, and coding. The block diagram of the proposed compression
method is shown in Figure 1.
3.1. Intra Prediction
Intra prediction [8] exploits spatial correlation within one image. In this process, the current block is
predicted by using the boundary pixel values of previously reconstructed neighboring blocks. As shown in
the Figure 2, the boundary pixel values of neighboring blocks as shown in shaded boxes are copied into the
current block pixels along a specified direction indicated by the mode. Eight directional modes and a DC
mode, which is almost same as the nine intra-prediction modes employed in H.264 standard [9].
(a) (b)
Figure 2. (a) Intra Prediction, (b) 8-Directional mode + DC Mode
In this proposed method, the image is divided into blocks of size 8 x 8. The nine intra-prediction
modes (0-8) are applied over each 8 x 8 block and residual error is calculated for each prediction mode. The
residual error is the difference between the pixel value of the current block and pixel value of the predicted
block. The best prediction mode for a current block is selected based on the minimum residual error. The
mode number M and the residual error block are transmitted for encoding.
3.2. Dictionary Training
The dictionary is trained off-line using prediction residual samples resulting from a wide variety of
images. We first divide different images into blocks of size 8 x 8 and then 9 intra- prediction modes (0-8) are
applied over each 8 x 8 block. The predicted block is subtracted from the current block to generate residual
blocks. A set of 8 × 8 prediction residual blocks for different modes are selected to train the dictionary. To
train the dictionary we employed K-SVD algorithm [14]. During dictionary training, in each iteration, K-
SVD algorithm updates the dictionary elements by optimizing minimization problem given in Equation(4).
K-SVD algorithm iteratively updates the dictionary elements to better fit the data and after certain iteration,
an updated dictionary is resulted. This updated dictionary is used for sparse representation of the residual
block during image coding.
3.3. Sparse Representation
OMP algorithm is employed for the sparse representation of residual block. OMP algorithm selects
the appropriate dictionary element to represent each residual image block. In each iteration, OMP selects the
best linear combination of dictionary elements by minimizing Equation (3). The same process is continued
and the algorithm terminates when the residual error of the reconstructed signal is equal to or less than a
specified value. However, the number of OMP iterations may not exceed T0. Once the algorithm terminates
the coefficients C and indices P of appropriate dictionary element is transmitted for encoding.
3.4. Coding
After sparse representation, the coefficients C and indices P of dictionary element, and the
prediction mode M are encoded [22]. The coefficients are uniformly quantized followed by entropy coding.

 ISSN: 2088-8708
1968
The indices are encoded with fixed length codes whose sizes are log2k, where k is the number of dictionary
elements. The prediction mode numbers are encoded with fixed length codes whose sizes are 4 bits. The
encoder of the proposed compression method is shown in Figure 1(a).
Since the dictionary is shared at both encoder and decoder, only intra- prediction mode number,
indices, and coefficients of the dictionary elements are transmitted to the decoder. The decoder generates the
residual block from the knowledge of dictionary, coefficients, and indices. The decoder then predicts the
block based on mode number and combines with the residual block to reconstruct the block. The decoder of
the proposed compression method is shown in Figure 1(b).
4. EXPERIMENTAL RESULTS
In this section, we conducted several experiments in order to evaluate the performance of the
proposed compression method. The proposed algorithm is applied over several images. The compression
efficiency and quality of the reconstructed image are compared with several other competitive image
compression techniques.
4.1. Intra Prediction
In the experiment, we used 100 images from the Berkeley segmentation database as our training set.
Nine modes of 8 x 8 intra prediction are applied and intra-prediction residual for each image is generated. A
set of 45000 blocks of size 8 x 8 are randomly selected from residual images to train a dictionary. We
selected 5000 residual blocks from each nine modes to form our learning set. Examples of intra-predicted
image and residual image are demonstrated in Figure 3. A random collection of such training blocks for
different mode is shown in Figure 4.
(a) Original Image (b) Predicted Image (c) Residual Image
Figure 3. Intra prediction of second image
(a) Mode 3 (b) Mode 6 (c) Mode 8
Figure 4. Training blocks

IJECE ISSN: 2088-8708 
1969
4.2. K-SVD Dictionary
We used 45000 residual blocks obtained from intra-prediction as our learning set. The K-SVD
algorithm is applied to train an over-complete dictionary of size 64 x 512, where the number of rows 64
represents the number of pixels in a block, and the number of columns 512 represents the number of
dictionary elements. In the training process, 100 number of K-SVD iterations was set as the primary stopping
criterion. Sparsity constraint (T0=4) was set as another stopping criterion. Example of a trained dictionary is
demonstrated in Figure 5.
Figure 5. Example of dictionary trained on 8x8 residual blocks
4.3. Image Compression
The K-SVD Dictionary resulting from prediction residuals is used for sparse representation. In the
experiment, the residual error (δ=0.2) and the number of OMP iterations equal to 4 was set as stopping
criterion of OMP algorithm. The OMP algorithm stops when the residual error is less than or equal to 0.2,
otherwise it stops after 4 iterations. In maximum, four dictionary elements are required to encode each
residual block.
(a) (b)
(c) (d)
Figure 6. Rate-distortion curves of different methods
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
22
24
26
28
30
32
34
36
38
40
Barbara
Bit-rate (bpp)
PSNR(dB)
JPEG
JPEG2000
Proposed Method
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
28
30
32
34
36
38
40
42
Lena
Bit-rate (bpp)
PSNR(dB)
JPEG
JPEG2000
Proposed Method
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
22
23
24
25
26
27
28
29
30
31
32
33
Baboon
Bit-rate (bpp)
PSNR(dB)
JPEG
JPEG2000
Proposed Method
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
28
30
32
34
36
38
40
Peppers
Bit-rate (bpp)
PSNR(dB)
JPEG
JPEG2000
Proposed Method

 ISSN: 2088-8708
1970
In order to evaluate our proposed compression method, we compare our method with JPEG and
JPEG2000. All experiments are performed using MATLAB. The performance of proposed method is
evaluated by taking different standard test images. The quantitative evaluation of our proposed method is
accomplished using two image quality metrics: PSNR and SSIM (Structural Similarity Metric) [23]. Figure 6
shows comparison of rate-distortion curve for 4 standard test images. Table 1 shows PSNR comparison for
different images at two different bit-rates. The proposed method yields around PSNR gain of 3 dB compared
to JPEG, and PSNR gain of 0.3 dB compared to JPEG 2000. The performance in terms of average PSNR and
SSIM are shown in Table 2. The results are averaged over 9 standard test images and best results are bolded.
Subjective assessment of one image is shown in Figure 7 at bit-rate 0.2. The results show that our proposed
method outperforms JPEG and JPEG2000 in terms of all quality metrics, including PSNR and SSIM.
Table 1. PSNR (dB) comparison of the proposed method with JPG and JPEG2000 for a set of 9 test images at
two different bit-rates. Best results are bolded
PSNR(dB) at bit-rate 0.2 bpp PSNR(dB) at bit-rate 1 bpp
Images JPEG JPEG2000 Proposed method JPEG JPEG2000 Proposed method
Barbara 23.81 27.29 27.60 34.46 37.15 37.52
Sailboat 26.56 29.12 29.53 34.44 36.81 37.01
Baboon 22.48 25.24 25.45 27.98 30.65 30.81
couple 25.72 28.48 28.80 34.25 36.79 37.10
Hill 26.78 29.88 30.21 33.62 36.42 36.84
Jet plane 29.56 31.92 32.28 39.65 41.88 42.20
Lena 29.43 33.02 33.46 37.88 40.40 40.82
Lighthouse 25.82 28.39 28.82 35.50 38.42 38.83
Peppers 29.98 32.49 32.88 37.55 38.38 38.72
Average 26.68 29.54 29.89 35.03 37.43 37.76
(a) Original image (b) JPEG (c) JPEG2000 (d) Proposed method
Figure 7. Subjective assessment
Table 2. Performance comparison in terms of two image quality metrics, PSNR (dB) and SSIM at six
different bit-rates. The results are averaged over 9 test images. Best results are bolded
Quality Metric JPEG JPEG2000 Proposed method JPEG JPEG2000 Proposed
method
Bit-rate 0.2 bpp Bit-rate 0.4 bpp
PSNR 26.68 29.54 29.89 28.78 31.82 32.14
SSIM 0.748 0.771 0.782 0.762 0.844 0.861
Bit-rate 0.6 bpp Bit-rate 0.8 bpp
PSNR 31.12 33.82 34.22 33.40 36.42 36.74
SSIM 0.824 0.906 0.915 0.874 0.921 0.931
Bit-rate 1 bpp Bit-rate 1.2 bpp
PSNR 35.03 37.43 37.76 36.43 38.55 38.87
SSIM 0.909 0.941 0.949 0.918 0.951 0.954
50 100 150 200 250 300 350 400 450 500
50
100
150
200
250
300
350
400
450
500
50 100 150 200 250 300 350 400 450 500
50
100
150
200
250
300
350
400
450
500
50 100 150 200 250 300 350 400 450 500
50
100
150
200
250
300
350
400
450
500
50 100 150 200 250 300 350 400 450 500
50
100
150
200
250
300
350
400
450
500

IJECE ISSN: 2088-8708 
1971
5. CONCLUSION
In this paper, we presented a dictionary based intra-prediction framework for image compression.
K-SVD algorithm is used in order to train a dictionary. We trained the dictionary with a variety of residual
blocks obtained from intra-prediction and then used this dictionary for sparse representation of an image.
OMP algorithm, fixed length coding, and entropy coding have employed for encoding. Different coding
results based on a set of test images are presented to compare the performance of the proposed method with
the existing methods. Experimental result shows that the proposed method outperforms JPEG and JPEG2000.
ACKNOWLEDGEMENTS
The authors are grateful to Dr. Gagan Rath for his support and guidance throughout this work.
REFERENCES
[1] Salomon, “Data Compression: The Complete Reference”, third ed., Springer, New York, 2004.
[2] W. Pennebaker, J. Mitchell, “JPEG Still Image Data Compression Standard”, Kluwer Academic Publishers, 1993.
[3] C. Christopoulos, A. Skodras, T. Ebrahimi, “The JPEG2000 Still Image Coding System: An Overview”, IEEE
Trans. Consum. Electron, 46 (2000) 1103-1127.
[4] A. Ennaciri, M. Erritali, M. Mabrouki, J. Bengourram, “Comparative Study of Wavelet Image Compression:
JPEG2000 Standart”, TELKOMNIKA Indonesian Journal of Electrical Engineering, vol. 16, no. 1, pp. 83-90, Oct.
2015.
[5] R. Chourasiya, A. Shrivastava, “A Study of Image Compression Based Transmission Algorithm using SPIHT for
Low Bit Rate Application”, Bulletin of Electrical Engineering and Informatics, vol. 2, no. 2, pp. 117-122, Jun.
2013.
[6] D. Narmadha, K. Gayathri, K. Thilagavathi, N. Sardar Basha, “An Optimal HSI Image Compression using DWT
and CP”, International Journal of Electrical and Computer Engineering, vol. 4, no. 3, pp. 411-421, Jun. 2014.
[7] M. Stirner, G. Seelmann, “Improved Redundancy Reduction for JPEG Files”, Proc. of Picture Coding Symposium
(PCS 2007), Lisbon, Portugal, Nov. 7-9, 2007.
[8] J. Yang, B. Yin, Y. Sun, N. Zhang, “A block- Matching based Intra Frame Prediction H.264/AVC”, ICME, 2006.
[9] T. Wiegand, G. J. Sullivan, G. Bjøntegaard, A. Luthra, “Overview of the H.264/AVC Video Coding Standard”,
IEEE Trans. Circuits Syst. Video Technology., vol. 13, no. 7, pp. 560-576, Jul. 2003.
[10] J. Wright, Y. Ma, J. Mairal, G. Sapiro, T. S. Huang, S. Yan, “Sparse Representation for Computer Vision and
Pattern Recognition”, Proc. IEEE, vol. 98, no. 6, pp. 1031–1044, Jun. 2010.
[11] R. Rubinstein, A. M. Bruckstein, M. Elad, “Dictionaries for Sparse Representation Modeling”, Proc. IEEE, vol.
98, no. 6, pp. 1045–1057, June 2010.
[12] J. Mairal, F. Bach, J. Ponce, and G. Sapiro, “Online dictionary learning for sparse coding,” in Proc. 26th Annu. Int.
Conf. Mach. Learning., 2009, pp. 689-696.
[13] O. Bryt, M. Elad, “Compression of Facial Images using the K-SVD Algorithm”, J. Vis. Commun. Image
Representation. vol. 19, no. 4, pp. 270-282, 2008.
[14] M. Aharon, M. Elad, A. Bruckstein, “K-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse
Representation,” IEEE Trans. Signal Processing., vol. 54, no. 11, pp. 4311-4322, Nov. 2006.
[15] J. Zepeda, C. Guillemot, E. Kijak, “Image Compression using Sparse Representations and the Iteration-tuned and
Aligned Dictionary”, IEEE J. Sel. Topics Signal Processing, vol. 5, no. 5, pp. 1061-1073, Sep. 2011.
[16] G. Shao, Y. Wu, A. Yong, X. Liu, T. Guo, “Fingerprint Compression based on Sparse Representation”, IEEE
Trans. Image Process., vol. 23, no. 2, pp. 489-501, Feb. 2014.
[17] Jing-Ya Zhu, Zhong-Yuan Wang, Rui Zhong, Shen-Ming Qu, “Dictionary based Surveillance Image
Compression”, J. Vis. Commun. Image R. vol. 31, no. 7, pp. 225-230. Jul. 2015.
[18] G. Rath, A. Sahoo, “A Comparative Study of some Greedy Pursuit Algorithms for Sparse Approximation”,
EUSIPCO2009, pp.398-402, Aug. 2011.
[19] S. Mallat, Z. Zhang, “Matching Pursuits with Time Frequency Dictionaries”, IEEE Transactions on Signal
Processing, vol. 41, no. 12, pp. 3397-3415, Dec. 1993.
[20] Y.C.Pati, R.Rezaiifar M, P.S.Krishnaprasad, “Orthogonal Matching pursuit: Recursive Function Approximation
with Application to Wavelet Decomposition”, In Proc. 27th Conf. on Sig. Sys. and Comp., vol.1, Nov. 1993.
[21] G. Rath, C. Guillemot, “On a Simple Derivation of the Complimentary Matching Pursuit”, Signal processing,
vol. 90, no. 2, pp.7 02-706, Feb. 2010.
[22] A. Bovik, Handbook of Image and Video Processing, 2nd ed.San Francisco, CA, USA: Academic, 2005.
[23] Z. Wang, A. C. Bovik, H. R. Sheikh, E. P. Simoncelli, “Image Quality Assessment: From Error Visibility to
Structural Similarity,” IEEE Trans. Image Processing., vol. 13, no. 4, pp. 600-612, Apr. 2004.

 ISSN: 2088-8708
1972
BIOGRAPHIES OF AUTHORS
Arabinda Sahoo. Currently, he is working as an Assistant Professor in the department of
Electronics & Communication Engineering at ITER, Siksha „O‟ Anusandhan University,
Bhubaneswar, Odisha, India. He has completed his B.E from Utkal University,
Bhubaneswar, Odisha, India and M.Tech from National Institute of Technology, Rourkela,
India. His research interests include signal and image processing, image compression and
sparse representations.
Pranati Das. Currently, she is working as an Associate Professor in the department of
Electrical Engineering at Indira Gandhi Institute of Technology, Sarang an Autonomous
Institute of Government of Odisha, India. She has completed her M.Tech & PhD from
Indian Institute of Technology, Kharagpur, India and Engineering graduation from
Sambalpur University, Odisha, India. She has published number of research and conference
papers in both national and international forum. Her area of research interests are Image
Processing and Signal Processing.

Dictionary based Image Compression via Sparse Representation

Recommended

More Related Content

What's hot (19)

Similar to Dictionary based Image Compression via Sparse Representation (20)

More from IJECEIAES (20)

Recently uploaded (20)

Dictionary based Image Compression via Sparse Representation