COMPRESSIVE SENSING BASED

IMAGE RECONSTRUCTION
Internal Guide External Guide
Prof. Jigna Patel Prof. Ketki Pathak
E.C .Department E. C. Department,
Dr. S. & S. S. Ghandhy GEC, Surat SCET, Surat.

Submitted By:-
SHERIN C ABRAHAM (150230727015)
M.E. (ELECTRONICS AND COMMUNICATION)
Wireless Communication Systems and Networks
Semester-3, Year 2016,
Dr. S. & S. S. Ghandhy Govt. Engineering College, Surat.
CONTENTS
• Introduction
• Applications
• Motivation
• Literature survey
• Objectives
• Problem statement
• Block Diagram
• Work Done
• Future works
• Conclusion
• References
Compressive sensing based image reconstruction
2
150230727015
INTRODUCTION – SIGNAL COMPRESSION
TECHNIQUE
• Shannon’s Nyquist sampling theorem
– A signal sampled at a Rate > 2* fmax - for fidelity of signal reconstruction.

• High bandwidth signals (HD videos) - High number of samples - Compression.

• Imaging systems (medical scanners) – High speed ADC required - Expensive.

N>>M
N M
sample compress transmit/store

M N
receive decompress

• Some small coefficients of DCT or WT coefficients can be discarded.

Compressive sensing based image reconstruction
3
150230727015
INTRODUCTION – COMPRESSIVE SENSING
• If the signal is “sparse” or compressible, then the acquired significant
information is sufficient to reconstruct.
• The Compressive Sensing (CS) theory asserts that one can recover certain
signals and images from far fewer samples or measurements.

“Can we not just directly measure the part that will not
end up being thrown away ?” Donoho

M N
X Compressed sensing Reconstruction

Compressive sensing based image reconstruction

4
150230727015
 INTRODUCTION – CS CONT..

• Compressive Sensing relies on

three principle
– Sparsity : which pertains to
the signal of interest.
– Incoherence : which pertains
to the sensing modality or
measurement taking.
Incoherence Non-linear
– Non-linear reconstruction reconstruction
(Optimisation): which
Sparse
pertains to the signal recovery. Signal Fig. 1: Basic model of CS [16]

Compressive sensing based image reconstruction 5

150230727015
SPARSITY
• The “information rate” of a continuous time signal
may be much smaller than suggested by its bandwidth.

• Most coefficients are small, and few large coefficients

capture most of the information as shown in Fig. 3.

• One can discard the small coefficients without much

perceptual loss.

• Sparse approximation in orthogonal basis is the heart

for efficient compression of images or signals
Fig. 2: Sparsity in
wavelet domain. [10]
Compressive sensing based image reconstruction
6
150230727015
SPARSITY DEPENDS ON DOMAIN

Fig. 3: Sparsity of a rich signal in another domain. [13]

Compressive sensing based image reconstruction
7
150230727015
INCOHERENCE
• The sparsity matrix Ψ and
measurement matrix Φ should be
highly incoherent for image recovery
from under sampled data
• Incoherence extends the duality
between time and frequency - objects
Sparse
signal
Incoherence having a sparse representation in a
domain must be spread out in other
Fig. 4: To show the role of
incoherence in image denoising.
domain
Noise like pattern without – A spike in the time domain is
incoherence is in the right side[16]
spread out in the frequency
domain.
Compressive sensing based image reconstruction
8
150230727015
COMPRESSIVE SENSING BASED IMAGE
RECONSTRUCTION

Nyquist Sampling Compressive Sensing

Sampling < (2*fmax)

>= (2*fmax)
frequency

Recovery Low pass filter Convex Optimisation,

algorithm Matching Pursuit

Compressive sensing based image reconstruction

9
150230727015
INTRODUCTION – CS MATHEMATICALLY

• x is original discrete signal which is n x 1 size in Rn .

• x is in time domain.
• x=Ψf
– f is the basis domain representation of signal x by the orthonormal basis Ψ
of size NxN.
– If signal x is K-sparse, it’s a linear combination of only K basis vectors.
• y = ФΨf = Acsf = Фx
– Acs = ФΨ is the sensing matrix.
– Ф is the measurement matrix. [11]
Compressive sensing based image reconstruction
10
150230727015
CS MATRIX REPRESENTATION

K < M << N y = Φx
Fig. 5: Matrix representation of CS. [13]

• The measurement process -- linear

Compressive sensing based image reconstruction
12
150230727015
CS- BASIC BLOCK DIAGRAM

f
x Orthonormal
basis Ψ
Measurement
matrix φ

Ψ* Inverse of Ψ Reconstruction
algorithm

Compressive sensing based image reconstruction

13
150230727015
EXAMPLE USING RICH SIGNAL

Compressive sensing based image reconstruction

14
150230727015
EXAMPLE
USING
S PA R S E
SIGNAL

Compressive sensing based image reconstruction

15
150230727015
EXAMPLE OF CS USING SPARSE SIGNAL

A signal -- K-sparse in one basis (sparsity basis) -- recovered from M

(cK) non adaptive linear projections onto a second basis (measurement
basis) that is incoherent with the first (where c is a small over-measuring
constant).

Compressive sensing based image reconstruction

16
150230727015
RESTRICTED ISOMETRY PROPERTY (RIP)

Compressive sensing based image reconstruction

17
150230727015
WHY CS IN MRI?

• Magnetic resonance imaging (MRI) is an essential medical imaging tool with

an inherently slow data acquisition process.
• Applying CS to MRI offers
– significant scan time reductions
– benefits for patients and health care economics.
• Medical imagery is naturally compressible by sparse coding in an appropriate
transform domain (e.g., by wavelet transform).
• MRI scanners naturally acquire encoded samples, rather than direct pixel
samples (e.g., in spatial-frequency encoding).

Compressive sensing based image reconstruction

18
150230727015
APPLICATIONS
• Medical Imaging
• Compressive Radio Detecting and Ranging.
• Analog to Information Converters
• Seismic Imaging
• CS in Camera

Fig. 6: CS application images. [17] [18] [19]

Compressive sensing based image reconstruction
19
150230727015
MOTIVATION

• The increasing demand for visual communication over Internet and wireless
networks.

• The amount of visual data is huge, while the bandwidth of transmission

channels is limited.

• MRI data collection is inherently slow. Decrease scan time is a challenge.

• The reconstructed image should be of good quality.

Compressive sensing based image reconstruction

20
150230727015
CS- BASIC BLOCK DIAGRAM

f
x Transform
Domain
Measurement
matrix φ

Inverse Reconstruction
transform algorithm

Compressive sensing based image reconstruction

21
150230727015
WAVELET TRANSFORM
• The input image X is fed into low pass filter (L) and high pass filter (H)
separately . The output of the two filters are the subsampled. The resulting low
pass sub band yL and high pass sub band yH

Fig. 7: Wavelet Transform and Inverse wavelet transform. [15]

• LL is a coarser version of the original image and it contains the approximation
information which is low frequency
• LH, HL, and HH are the high frequency sub bands containing the detail
information.
Compressive sensing based image reconstruction
22
150230727015
 WAVELET
TRANSFORM
CONT..

Fig. 8: Multi level wavelet transform[15]

The 2D implementation of the discrete

wavelet transform (DWT) is to perform the
one dimensional DWT in row direction and it
is followed by a one dimensional DWT in
column direction. Fig. 9: Wavelet coefficients of an image. [10]
Compressive sensing based image reconstruction
150230727015 23
CS- BASIC BLOCK DIAGRAM

f
x Transform
Domain
Measurement
matrix φ

Inverse Reconstruction
transform algorithm

Compressive sensing based image reconstruction

24
150230727015
IMAGE RECONSTRUCTION TECHNIQUE

Compressive sensing based image reconstruction

25
150230727015
INTRA – PREDICTION

• Fig. 10: Different modes of intra prediction. [2]

Compressive sensing based image reconstruction

26
150230727015
LITERATURE REVIEW

Compressive sensing based image reconstruction

27
150230727015
PAPER 1
• Title: Robust Image Coding Based Upon Compressive Sensing
• Authors: Chenwei Deng, Weisi Lin, Bu-sung Lee, and Chiew Tong Lau
• Publication & Year: IEEE Transactions on Multimedia, vol. 14, no.2, April
2014
• Problem Identified in the paper:
Existing CS based algorithms are not robust in lossy channels. Rate-
distortion performance should be high.
• Measuring Parameters
PSNR for different measurement matrices, for different values of M
value.
• Research Gaps
Sparsity tuning can be done.
Compressive sensing based image reconstruction
28
150230727015
REVIEW
• CS-based algorithm by considering the intra- and inter-similarity among DWT
coefficients.
• Two random CS sensing matrices – more measurements of low-frequency
components.
• In the existing wavelet-based CS schemes, DWT is mainly applied for sparse
representation and the correlation of DWT coefficients has not been fully
exploited yet.
• To improve compressibility (and therefore the coding efficiency) without
sacrificing the robustness, the statistics of DWT coefficients has been investigated.
• After multi-scale DWT, the low- and high-frequency sub bands are re-sampled
separately.

Compressive sensing based image reconstruction

29
150230727015
PAPER 2

• Title: Compressed Sensing Image Reconstruction Using Intra Prediction

• Authors: Yun Song, Wei Cao, Yanfei Shen and Gaobo Yang
• Publication & Year: Science direct, Neurocomputing, 2015
• Problem Identified:
More sparsity signal x is, more high quality the reconstruction signal
• Measuring Parameters
PSNR for different measurement matrices, for different values of M
value.
• Research Gaps
Eight directional modes use for intra-prediction

Compressive sensing based image reconstruction

30
150230727015
REVIEW

• The sparsity level of prediction residual is higher than its original image block.

• Post processing to suppress the blocking artefacts.

• CS image reconstruction algorithm using intra prediction method based on

block-based CS image framework.

• The current reconstruction block is firstly predicted by its surrounding

reconstructed pixels, and then its prediction residual will be reconstructed

• Because the sparsity level of prediction residual is higher than its original
image block, the performance will be more better.
Compressive sensing based image reconstruction
31
150230727015
PAPER 3
• Title: Image Reconstruction Based on The Improved Compressive Sensing Algorithm
• Authors: Xiumeni Li and Guoan Bi
• Publication & Year: IEEE International Conference on Digital Signal Processing
(DSP), 2015
• Problem Identified:
Low frequency coefficients are not sparse, multiplying with measurement
matrix, the coherence will be disrupted
• Measuring Parameters
PSNR for different measurement matrices, for different values of M value.
• Research Gaps
Low frequency coefficients have coherence nature which can be utilized by intra
prediction
Compressive sensing based image reconstruction
32
150230727015
REVIEW

• The high-frequency coefficients are sparse

• The low frequency coefficients are not sparse

• The low-frequency coefficients contain most energy of the image

• Only measure the high-frequency sub band coefficients, and keep the low-
frequency sub-band coefficients unchanged.

• Use the reconstruction algorithms to reconstruct the three high-frequency

coefficients.

Compressive sensing based image reconstruction

33
150230727015
PAPER 4
• Title: Sparsity Fine Tuning in Wavelet Domain with Application to Compressive Image
Reconstruction
• Authors: Weisheng Dong, Xiaolin Wu, and Guangming Shi
• Publication & Year: IEEE Transactions on Image Processing, vol. 23, no.12, December
2014
• Problem Identified:
High frequency coefficients have non-zero bias. Removing the bias can make
wavelet image representations sparser
• Measuring Parameters
PSNR for different measurement matrices, for different values of M value.
• Research Gaps
Low frequency components are not processed
Compressive sensing based image reconstruction
34
150230727015
REVIEW

• More sparsity will improvise performance.

• High frequency wavelet coefficients non – zero mean distributions are made to
zero, thus improve sparsity of the image.

• It is possible to make a wavelet image representation sparser by estimating and

removing the conditional expectation μi = E{P(αi|Ci)} from αi.

• The resulting conditional PDF P(αi −μi|Ci) is centered at zero, (αi−μi) is more
likely to be zero or very small than αi, meaning that the bias-removed wavelet
representation becomes sparser.
Compressive sensing based image reconstruction
35
150230727015
PAPER 5
• Title: A New Wavelet Based Efficient Image Compression Algorithm Using
Compressive Sensing
• Authors: Muhammad Ali Qureshi and M. Deriche
• Publication & Year: Springer Science Multimedia tools and applications,
March 2015
• Problem Identified:
Measurement matrix is fixed and not depend on signal
• Measuring Parameters
PSNR for different measurement matrices, for different values of M
value.
• Research Gaps
More sparsity, Gaussian matrix will perform well.
Compressive sensing based image reconstruction
36
150230727015
REVIEW

• An image dependent measurement matrix where more weights are assigned to

low-frequency (high energy) components.

• A normalized Gaussian random measurement matrix based on the energies of

the wavelet coefficients.

• Rearrange wavelet coefficients in structured manner an optimal rearrangement

of the wavelet coefficients to form sparse vectors for CS.

• CS-based compression algorithm using this new type of sparse vector.

Compressive sensing based image reconstruction

37
150230727015
LITERATURE SURVEY SUMMARY

Compressive sensing based image reconstruction

38
150230727015
 Problem Autho Publicatio
No Paper Summary Gaps Year
identified rs n
Existing CS CS-based algorithm by
based algorithms considering the intra- and Chen IEEE
Robust are not robust in inter-similarity among DWT wei Transactio
Sparsity tuning April,
1 Image lossy channels. coefficients. Two random CS Den ns on
can be done. 2014
coding Rate- distortion sensing matrices – more and Multimedi
performance measurements of low- et.al a
should be high. frequency components

Image More sparsity The sparsity level of

Yun ScienceDi
reconstructi signal x is, more prediction residual is higher Eight directional Octob
Song rect
2 on using high quality the than its original image block. modes use for er,
and NeuroCo
intra reconstruction Post processing to suppress intra-prediction 2014
et.al mputing
prediction signal the blocking artifacts
Low frequency Low frequency
Xium IEEE
Improved coefficients are Low frequency keep coefficients have
eni Li Internation
Compressiv not sparse, taking unchanged. Use the same coherence nature
3 and al 2015
e Sensing measurements, coefficients for which can be
Guoa Conferenc
Algorithm the coherence reconstruction utilized by intra
n Bi e on DSP
will be disrupted prediction 39
 Publicatio
No Paper Problem identified Summary Gaps Authors Year
n
High frequency The conditional PDF
IEEE
Sparsity coefficients have non- P(αi −μi|Ci) is centred Weishe Dec
Low frequency Transactio
tuning in zero bias. Removing at zero. Improve ng emb
4 components are ns on
Wavelet the bias can make performance of Dong er,
not processed Image
Domain wavelet image structures such as and et.al 2014
Processing
representations sparser edges and textures.
Image dependent Springer
Wavelet
measurement matrix, M. A. Science
Based Measurement matrix is More sparsity, Mar
energy based. Qureshi Multimedi
5 Image fixed and not depend Gaussian matrix ch,
Rearrange wavelet and M. a tools and
compressio on signal will perform well. 2015
coefficients in Deriche applicatio
n Algorithm
structured manner. ns

High gradient
Sparse nature, Developing
amplitudes and rapid IEEE
incoherence of MRI sparsity. Improve Michael Mar
switching produce Signal
6 CS - MRI images. Applications speed of Lustig ch,
nerve stimulation. Processing
of CS in MRI are also reconstruction and et.al 2008
MRI data are Magazine
explained. algorithms
redundant 40
No Paper Problem identified Summary Gaps Authors Publication Year

Image Sabbise Green

Well designed sensing
acquisitio Add sparsity by OMP is not tti Computing
matrix can increase
n and sparsity-inducing compared with Ravindr Communicati
7 the incoherence and 2014
Reconstru matrices . OMP is the L1 minimization anath on and
reduce the
ction efficient algorithm. technique and et. Electrical
reconstruction error.
analysis al Engineering

Chen
Science
Sparsity in wavelet Chen
Wavelet Wavelet tree structure- Slow Direct
domain or gradient and July,
8 Structure fewer data only computation due Magnetic
domain only Junzho 2014
CS MRI required. to tree structure Resonance
exploited. u
Imaging
Huang

Sparsity, enables the Complexity and IEEE Journal

signal to be stored in minimum measurement of
CS - Saad
9 few samples and requirement of CS communicati 2013
survey Qaisar
subsequently be reconstruction ons and
recovered accurately algorithm are analysed networks 41
OBJECTIVES

• Reduce bandwidth using wavelet transform

• Reduce the processing time using Block-based CS method.

• Add enough sparsity by sparsity tuning and intra-prediction methods.

• Use good measurement matrices separately for low frequency components and
high frequency components.

• Use proper reconstruction algorithm.

Compressive sensing based image reconstruction

42
150230727015
PROBLEM STATEMENT

• To optimise compressive based algorithm for better visual quality for medical
image application.

Compressive sensing based image reconstruction

43
150230727015
 BLOCK DIAGRAM – BASE PAPER
Original Image Reconstructed Image

Encoder DWT (ψ) IDWT Decoder

Low- High- CS recovery - 1 CS recovery - 2

frequency XL frequency XH
Block Regrouping φL Regrouping φH

Side Information
partition Y Y
Measurements Measurements
Gabor sampling (φL) sampling (φH) Bit- N Bit- N
filtering loss? loss?
YL YH
Similar N Quantization
output?
De-quantization
Y Packeting
Intra- Un-packet
prediction Entropy Coding
Entropy decoding
Side info.
Channel
Compressive sensing based image reconstruction
Block diagram from base paper[1] 150230727015
44
 IMPROVEMENT & SUGGESTIONS
Original Image Channel
(256x256)

2D DWT coefficients after Decoding Decoding

sparsity fine tuning

CS Recovery
Block Low frequency High frequency using L1 CS Recovery
partition components components minimization using OMP

Measurements & Measurements &

Prediction Encoding using Encoding using IDWT
and filtering Hadamard matrix Gaussian matrix
Reconstructed
Channel Image

Compressive sensing based image reconstruction

45
150230727015
MEASURING PARAMETERS

Compressive sensing based image reconstruction

46
150230727015
WORK DONE

Compressive sensing based image reconstruction

47
150230727015
IMAGES USED FOR SIMULATIONS

32x32 64x64 128x128 256x256 BACK

Fig. 11: Images of size 32x32, 64x64, 128x128, 256x256. [17]

Compressive sensing based image reconstruction

48
150230727015
RESULTS

32x32 64x64 128x128 256x256 512x512

L1 & PINV - 50.08 42.74 33.31 31.46

PSNR OMP - - 32.38 20.77 16.99

L1 - 50.08 42.74 - -

L1 & PINV 0 0.6382 3.45 30.28 46.36

MSE OMP 0 0 610 542.94 1991.1

L1 0 0.6382 3.45 - -
In OMP, each image undergoes 50 iterations.
Compressive sensing based image reconstruction
49
150230727015
MATLAB OUTPUT IMAGES

BACK

Fig. 12 : Output of 128x128 image using various techniques

Compressive sensing based image reconstruction
50
150230727015
DIFFERENT IMAGES

Fig. 13: Different medical images [17]

Compressive sensing based image reconstruction

51
150230727015
RESULTS FOR DIFFERENT IMAGES

Medical Medical Medical Medical

image1 image2 image3 image4

L1 & PINV 43.83 54.85 31.76 30.11

PSNR
OMP 15.28 18.21 19.77 14.23

L1 & PINV 2.699 0.212 43.27 63.36

MSE
OMP 1971.3 981.0 684.86 2455.1

In OMP, each image undergoes 50 iterations.

Compressive sensing based image reconstruction

52
150230727015
PSNR FOR DIFFERENT ITERATIONS IN OMP
PSNR PSNR PSNR PSNR
PSNR
Iterations using using using using
using 64x64
32x32 128x128 256x256 512x512
5 17.87 15.76 14.53 13.5 12.96

10 23.37 18.57 15.86 14.6 13.66

15 36.67 21.05 17.25 15.34 14.14

25 - 30.39 20.25 16.84 15.1

50 - - 32.38 20.77 16.99
100 - - - 33.88 21.12
Compressive sensing based image reconstruction
53
150230727015
PSNR Vs. SIZE OF AN IMAGE

Fig. 14: PSNR of different sizes of an image for 15 iterations in OMP reconstruction algorithm.
Compressive sensing based image reconstruction
54
150230727015
MATLAB OUTPUT IMAGES

Fig. 15 : Output of 128x128 image using different iterations of OMP

Compressive sensing based image reconstruction
55
150230727015
PSNR VS ITERATIONS

Fig. 16: PSNR for different iterations in OMP reconstruction algorithm.

Compressive sensing based image reconstruction
56
150230727015
INFERENCES DRAWN
• Low frequency components and high frequency components (after wavelet
transform) should be processed separately.

• Low frequency components are the approximation of the image (not sparse)
and high frequency components are details (sparse) of the image.

• Hadamard matrix will give better image quality. For n x n matrix, HHT = n In .

• L1 minimisation technique is better than OMP for large size images, and vice
versa for small images.

• More iterations, better quality image can be reconstructed using OMP.

Compressive sensing based image reconstruction
57
150230727015
CONCLUSION
• Compressive sensing measures a relatively small number of “random” linear
combinations of the signal value.

• Sparsity, incoherence, non linear reconstruction are three main requirements

of Compressive sensing.

• Wavelet transform is proven sparsity domain for many signals.

• Low frequency band which have approximate information can be further

processed by applying intra-prediction.

• In Medical field, CS will help in less number of samples, less radiation for
patients.
Compressive sensing based image reconstruction
58
150230727015
PLANNING FOR NEXT SEMESTER

• More sparsity can be introduced in the image.

• Need to do Intra-prediction for low frequency components.

• Need to verify other CS algorithms to get better image in visual quality in

adverse conditions.

Compressive sensing based image reconstruction

59
150230727015
 REFERENCES
1. Chenwei Deng, Weisi Lin, Bu-sung Lee, and Chiew Tong Lau, "Robust Image Coding Based
Upon Compressive Sensing”, IEEE Transactions on Multimedia, vol. 14, no.2, April 2014.
2. Yun Song, Wei Cao, Yanfei Shen and Gaobo Yang, “Compressed sensing image reconstruction
using intra prediction”, Science direct, Neurocomputing, 2015
3. Xiumeni Li and Guoan Bi, “Image reconstruction based on the improved compressive sensing
algorithm”, IEEE International Conference on Digital Signal Processing (DSP), 2015
4. Weisheng Dong, Xiaolin Wu, and Guangming Shi, "Sparsity Fine Tuning in Wavelet Domain
With Application to Compressive Image Reconstruction”, IEEE Transactions on Image
Processing , vol. 23, no.12, December 2014.
5. Muhammad Ali Qureshi and M. Deriche, “A new wavelet based efficient image compression
algorithm using compressive sensing”, Springer Science Multimedia tools and applications,
March 2015
6. Michael Lustig and et.al, “ Compressed sensing MRI”, IEEE signal processing magazine,
March, 2008
Compressive sensing based image reconstruction
60
150230727015
 REFERENCES CONT..
7. Sabbisetti Ravindranath and et. al, “Compressive Sensing based Image acquisition and
Reconstruction analysis” Green Computing Communication and Electrical Engineering
(ICGCCEE), 2014 International Conference
8. Chen Chen and Junzhou Huang, “Exploiting the wavelet structure in compressed sensing
MRI”, Science Direct Magnetic Resonance Imaging, July 2014.
9. Saad Qaisar ande et.al, “Compressive sensing: From theory to applications, a survey”, IEEE
Journal of communications and networks, vol.15, no.5, October 2013
10. Emmanuel J. Candès and Michael B. Wakin, “An introduction to compressive sensing”, IEEE
Signal Processing Magazine, March 2008
11. S. Spurthi and Parnasree Chakraborty, “Block based compressed sensing algorithm for medical
image compression”, International Journal of Engineering and Computer Science, vol. 05,
issue 5, May 2016
12. Sumit Budhiraja and Dipti Bhatnagar, “Wavelet based compressive sensing techniques for
image compression”, International journal of Engineering Research and Applications, 2011.
Compressive sensing based image reconstruction
61
150230727015
 REFERENCES CONT..
Websites:
13. http://dsp.rice.edu/cs
14. http://dsp.rice.edu/sites/dsp.rice.edu/files/cs/baraniukCSlecture07.pdf
15. https://en.wikipedia.org/wiki/Compressed_sensing
16. http://www.slideshare.net/thomasstefani169/compressive-sensing-basics-medical-imaging-mri
Image Database:
17. Civil Hospital, Surat
18. https://www.ncm-c.org/services/diagnostic-imaging/magnetic-resonance-imaging-mri/
19. http://www.healthhype.com/diagnostic-imaging-of-the-abdomen-x-ray-ultrasound-ct-mri.html

Compressive sensing based image reconstruction

62
150230727015
THE END

Compressive sensing based image reconstruction

63
150230727015