eurocon2009 khryashchev

32
1 EUROCON 2009, May 21, Saint- Peterburg NO-REFERENCE QUALITY ASSESSMENT OF WAVELET-COMPRESSED IMAGES Vladimir Khryashchev Yaroslavl State University Russia

Upload: khryashchev

Post on 27-Jan-2015

104 views

Category:

Technology


0 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Eurocon2009 Khryashchev

1EUROCON 2009, May 21, Saint-Peterburg

NO-REFERENCE QUALITY ASSESSMENT OF WAVELET-COMPRESSED IMAGES

Vladimir Khryashchev

Yaroslavl State UniversityRussia

Page 2: Eurocon2009 Khryashchev

2EUROCON 2009, May 21, Saint-Peterburg

Outline

• Image compression: JPEG vs. JPEG2000

• Image quality assessment: from PSNR to No-Reference estimation

• Algorithm description

• Results

• Conclusions

Page 3: Eurocon2009 Khryashchev

3

Image compression standard history

• JPEG – 1994 ISO/IEC 10918-1, ITU-T.Rec.T.81

• JPEG 2000 – December 2000ISO/IEC 15444-1

The experts decided to depart from the block based DCT coding used byexisting JPEG and MPEG standards in favor of wavelet based compressionwhich delivered better quality.

EUROCON 2009, May 21, Saint-Peterburg

Page 4: Eurocon2009 Khryashchev

4

JPEG Artifacts

Original Lenna Image Lenna with JPEG compression (ratio K=40)

8x8 block structure !!!

EUROCON 2009, May 21, Saint-Peterburg

Page 5: Eurocon2009 Khryashchev

5

JPEG2000 Artifacts

Original Lenna Image

Blurring !!!

Lenna with JPEG2000 compression (ratio K=80)

EUROCON 2009, May 21, Saint-Peterburg

Page 6: Eurocon2009 Khryashchev

6

JPEG2000 Artifacts

Original Lenna Image

Ringing !!!

Lenna with JPEG2000 compression (ratio K=80)

EUROCON 2009, May 21, Saint-Peterburg

Page 7: Eurocon2009 Khryashchev

7EUROCON 2009, May 21, Saint-Peterburg

JPEG vs JPEG2000 Comparison (Classical)

0 10 20 30 40 50 6024

26

28

30

32

34

36

38

40

K

PS

NR

, dB

JPEG2000JPEG

JPEG without deblocking, JPEG 2000 Kakadu v. 6.0

Page 8: Eurocon2009 Khryashchev

8EUROCON 2009, May 21, Saint-Peterburg

JPEG vs JPEG2000 Comparison (Classical)

JPEG 2000 delivers about 20% better compression than JPEG. And, at more extreme compression ratios, JPEG 2000 delivers significantly better quality.

Another key benefit of JPEG 2000 is that it supports both losslessand lossy compression in a single codec – a very desirable feature incertain applications such as medical imaging.

Page 9: Eurocon2009 Khryashchev

9EUROCON 2009, May 21, Saint-Peterburg

JPEG vs JPEG2000 Comparison (Modern)

JPEG with deblocking, JPEG 2000, SPIHT, AGU (DCT 32x32)

N.N. Ponomarenko, K.O. Egiazarian, V.V. Lukin, J.T. Astola, High-Quality DCT-Based Image Compression

Using Partition Schemes, IEEE Signal Processing Letters , Vol. 14, Issue 2, February, 2007, pp. 105-108.

http://ponomarenko.info/agu.htm

Page 10: Eurocon2009 Khryashchev

10EUROCON 2009, May 21, Saint-Peterburg

Motion JPEG2000 Standard for Digital Cinema

In Motion JPEG 2000 each frame is coded individually. Intra-frame coding allows for random access and reduced complexity. In MPEG the encoding uses a series of frames – inter-frame coding. This allows for improved compression efficiency but the coding technique is more complex.

Page 11: Eurocon2009 Khryashchev

11EUROCON 2009, May 21, Saint-Peterburg

From MSE and PSNR to modern image quality assessment

Z. Wang and A.C. Bovik, "Mean squared error: Love it or leave it? - A new look at signal fidelity measures", IEEE Signal Processing Magazine, Vol: 26 No: 1, Page(s): 98-117, January 2009.

N. Ponomarenko, F. Battisti, K. Egiazarian, J. Astola, V. Lukin "Metrics performance comparison for color image database", Fourth international workshop on video processing and quality metrics for consumer electronics, Scottsdale, Arizona, USA. Jan. 14-16, 2009, 6 p.

Introduction paper

Detailed analysis for 18 metricsand 17 types of distortions

More than 20 new full-reference metrics

Page 12: Eurocon2009 Khryashchev

12EUROCON 2009, May 21, Saint-Peterburg

The problem formulation

create a no-reference (working without original image) metrics for estimation quality of JPEG2000 compression images

Quality ?

Page 13: Eurocon2009 Khryashchev

13EUROCON 2009, May 21, Saint-Peterburg

No-reference Algorithm for JPEG2000

Based on statistical model of the wavelet coefficient’s magnitude proposed by Simoncelli (1997) and Bovik at al (2002) algorithm of quality estimation.

Page 14: Eurocon2009 Khryashchev

14EUROCON 2009, May 21, Saint-Peterburg

Joint histograms for one subband of Lenna image

original K=3,29

K=10,26 K=42,11

Page 15: Eurocon2009 Khryashchev

15EUROCON 2009, May 21, Saint-Peterburg

Partition of the (P, C) space into quadrants

Page 16: Eurocon2009 Khryashchev

16EUROCON 2009, May 21, Saint-Peterburg

Quality estimation calculation

Page 17: Eurocon2009 Khryashchev

17EUROCON 2009, May 21, Saint-Peterburg

Quality estimation calculation

No-reference image quality estimation (NIQE2000)from [0, 100] interval

where the weights ω are learned by minimizing quality prediction error over the training set

Page 18: Eurocon2009 Khryashchev

18EUROCON 2009, May 21, Saint-Peterburg

MOS estimation for JPEG2000 images

Visual experiment in Yaroslavl State University10 test images, 7 compression ratio for JPEG2000, 100 peoples.

ITU-R BT.500-11 methodology for DMOS calculations

Page 19: Eurocon2009 Khryashchev

19EUROCON 2009, May 21, Saint-Peterburg

Subband labels

1H

1V 23H

4

5H

5V 6

7H

7V 8

Page 20: Eurocon2009 Khryashchev

20EUROCON 2009, May 21, Saint-Peterburg

Subband visualization

Page 21: Eurocon2009 Khryashchev

21EUROCON 2009, May 21, Saint-Peterburg

Parameters optimization

Obtained algorithm parameters’ values for finest and second-finest scales are shown in this table. Last two scales of the wavelet transformation were not used since quantization process has little effect on them.

Page 22: Eurocon2009 Khryashchev

22EUROCON 2009, May 21, Saint-Peterburg

Visual Results

Original Image K=17, PSNR=36,61 dB,UQI=0,60, NIQE2000=60,12

Page 23: Eurocon2009 Khryashchev

23EUROCON 2009, May 21, Saint-Peterburg

Visual Results

K=100, PSNR=28,97 dB,UQI=0,37, NIQE2000=31,61

K=50, PSNR=31,96 dB,UQI=0,46, NIQE2000=35,62

Page 24: Eurocon2009 Khryashchev

24EUROCON 2009, May 21, Saint-Peterburg

NIQE2000 with PSNR

Page 25: Eurocon2009 Khryashchev

25EUROCON 2009, May 21, Saint-Peterburg

Correlation between MOS and objective estimations

Page 26: Eurocon2009 Khryashchev

26EUROCON 2009, May 21, Saint-Peterburg

Conclusions

NIQE2000 criteria can be successfully used for quality assessment of images compressed with JPEG2000 algorithm.

The correlations between NIQE2000 and MOS is close to PSNR-MOS, but NIQE2000 is no-reference estimation with very important for practical tasks.

Another no-reference quality assessment algorithms for JPEG2000 can be done by estimation ringing and blurring artifacts separately and after combining them to the new quality metrics.

Page 27: Eurocon2009 Khryashchev

27EUROCON 2009, May 21, Saint-Peterburg

Real-Life JPEG2000 Yaroslavl Image Quality Estimation

NIQE = 100

Page 28: Eurocon2009 Khryashchev

28EUROCON 2009, May 21, Saint-Peterburg

Real-Life JPEG2000 Yaroslavl Image Quality Estimation

NIQE = 100

Page 29: Eurocon2009 Khryashchev

29EUROCON 2009, May 21, Saint-Peterburg

Real-Life JPEG2000 Yaroslavl Image Quality Estimation

NIQE = 100

Page 30: Eurocon2009 Khryashchev

30EUROCON 2009, May 21, Saint-Peterburg

Real-Life JPEG2000 Yaroslavl Image Quality Estimation

NIQE = 100

Page 31: Eurocon2009 Khryashchev

31EUROCON 2009, May 21, Saint-Peterburg

Yaroslavl 1000 anniversary (1010-2010)

You are Welcome !!!

Page 32: Eurocon2009 Khryashchev

32EUROCON 2009, May 21, Saint-Peterburg

Thank you for your attention !!!