You are on page 1of 24

Image Compression:

Comparative Analysis of Basic Algorithms

Yevgeniya Sulema (Ukraine)


Samira Ebrahimi Kahou (Iran)

National Technical University of Ukraine


“Kyiv Polytechnic Institute”
sulema@scs.ntu-kpi.kiev.ua
samira_ebrahimi@hotmail.com
Outline
 Existing compression methods and
classification
 Criteria
 How to choose image set for testing
 Realizing algorithms
 Getting numerical values on chosen criteria
 Verifying results obtained from test
 Analysis and conclusion
Compression algorithms-Classification

5 Main Classification Types chosen.

By data type :

 General algorithms
 Algorithms for audio-compression
 Algorithms for image-compression
 Algorithms for video-compression
Compression algorithms-Classification (..2)

By data source :

 Dynamic
 Static

By redundancy type :

 Statistical redundancy reduction


 Spatial redundancy reduction
Compression algorithms-Classification (..3)
By restoring the original dataset:

 Lossless
 Lossy

By computational approach :

 Statistical
 Dictionary
 Transformation based
 Hybrid
Classes of Images
 Business graphics (schemes, diagrams, charts)
 Pictures created in graphic editors (photoshop)
 Photorealistic images (photos, textures)

Coefficient of correlation can be used between


an analyzed (test) image and an etalon image to
classify images :

cov I A , I E 

D( I A )  D( I E )
Sample images

 Image with two monochrome areas


 Image with large monochrome fields
 Gradient image
 Image with small monochrome fields
Correlation coefficients
(Sample images)

0.5 0.437366167

0.4

0.3 0.239146336

0.2
0.062622429
0.1

-0.1

-0.2

-0.3
-0.300682476
-0.4
image 1 image 2 image 3 image 4
Criteria
1. Compression ratio
2. Time of compression
3. Time of decompression
4. Peak signal-to-noise ratio (PSNR)
 max( I ) 
  20  log10  
  
 
MSE : Mean Squared Error
n m
1
 pi, j  pi, j
2

nm
i 1 j 1

5. Coefficient of correlation between original


and decompressed image
Matlab image processing Toolbox
Why Matlab?

It provides a comprehensive set of reference-


standard algorithms.
The software is a collection of functions that
extend the capability of the MATLAB.
The toolbox supports a wide range of image
processing operations.
Most toolbox functions are written in the open
MATLAB language, giving us the ability to
inspect the algorithms, modify the source code.
Algorithms:

Lossless : Lossy :
 LZW  JPEG
 LZ77 (Coarse and Fine)
 Wavelet
 Huffman
(Daubechies, Coiflets, Symlets,
 Adaptive Huffman Discrete Meyer wavelet,
Biorthogonal, Reverse
 Shannon-Fano Biorthogonal)
 Arithmetic  SPIHT
 Fractal
Lossless
Time of compression
102.6
100 91.8
87.0
78.8
80
63.2
59.6
60

40
25.7

20 12.2 11.8
4.8 5.8 6.6
1.9 3.7
0.7 0.9

0
Image 1 Image 2 Image 3 Image 4

LZ77 Huffman ShannonFano Arithmetic


Lossless
Time of decompression
113.8

100

80
63.8

60 51.6

40
20.2
15.9
20 10.8
0.8 0.6 1.1 2.2 1.9 2.6 0.8 2.8 3.5 4.0

0
Image 1 Image 2 Image 3 Image 4

LZ77 Huffman ShannonFano Arithmetic


Lossless
Compression ratio
10 9.0
9
8.0 8.0
8 7.3 7.1

3
1.4 1.5 1.4 1.5
2 1.1 1.1 1.2 1.0 1.0 1.0
0.7
1

0
Image 1 Image 2 Image 3 Image 4

LZ77 Huffman ShannonFano Arithmetic


Lossless Algorithm Observation
Dictionary Based Algorithms most
Effective

LZ77 – prime example from our research

Minimal Time for Compression


Minimal Time for Decompression
High Compression Ratio
Lossy
Time of compression
157.0 154.4 153.6 154.0

140

120

100

80

60

40

20
0.1 0.1 0.7 0.1 0.2 0.9 0.1 0.1 0.7 0.1 0.4 1.1
0
Test1 Test2 Test3 Test4

JPEG Coarse JPEG Fine SPIHT Fractal


Lossy
Time of decompression
1
0.9
0.9

0.8

0.7 0.6
0.6 0.6 0.6
0.5 0.5 0.6 0.5
0.6
0.5 0.5 0.5 0.5
0.5

0.4 0.3
0.3
0.3 0.2
0.2

0.1

0
Test1 Test2 Test3 Test4

JPEG Coarse JPEG Fine SPIHT Fractal


Lossy
Compression ratio
19.2 18.8
20
17.6
18

16

14 12.3
11.9
12

10
8
4.8
6 4.0 4.0 4.0 4.0 4.0 4.0 4.0 4.0
3.5
4
1.2
2

0
Test1 Test2 Test3 Test4

JPEG Coarse JPEG Fine SPIHT Fractal


Lossy
Correlation coefficient
1.0 1.0 1.0 1.0 0.9 1.0 1.0 1.0 1.0 1.0 1.0 1.0
1

0.9
0.8
0.7
0.7 0.6
0.6
0.5
0.5
0.4
0.3
0.2
0.1 0.0
0
Test1 Test2 Test3 Test4

JPEG Coarse JPEG Fine SPIHT Fractal


Lossy
PSNR
99.2
100

90

80

70

60 53.5
48.1 48.1
50 41.1
37.7 38.1 37.8
35.8
40 32.7 31.0 29.8
27.6 26.2 27.7
30 21.7

20

10

0
Test1 Test2 Test3 Test4

JPEG Coarse JPEG Fine SPIHT Fractal


Lossy Algorithm Observations
Fractal Algorithm not practical.

All remaining algorithms are Hybrid

Combination of procedures can result in


increased quality.
Conclusion
Our research allows us to draw 3 main conclusions:

 The selection of the proper compression


algorithm for each image class should be made

 Hybrid algorithms, JPEG, can be modified in


order to achieve better result

 Combination of a dictionary and transforms


most promising.
Thank YOU!

Questions…???

samira_ebrahimi@hotmail.com

You might also like