A Universal Image Quality Index

1
A Universal Image Quality Index

• Author Zhou Wang, Student Member, IEEE,
• and Alan C. Bovik, Fellow, IEEE
• Source IEEE SIGNAL PROCESSING
• LETTERS, VOL. 9, NO. 3, MARCH 2002
• Advisor Chin-Chen Chang
• Speaker Yi-Chang Liu
• Date 2004/4/21

2
Outline

• Introduction
• Definition of The New Quality Index
• Application to Images
• Conclusion and Discussion

3
Introduction

• Two classes of objective quality or distortion
  approaches
• MathematicallyMean Squared Error (MSE)?Peak
  Signal to Noise Ratio (PSNR)
• Human Visual System (HVS)

4
Introduction (con.)

• Why mathematically defined measures?
• Easy to calculate
• Usually have low computational complexity
• Independent of viewing
• Individual observers

5
Introduction (con.)

• Universal
• The images being tested
• The viewing conditions
• The individual observers

6
Definition of The New Quality Index
(The original image signals)
(The change image signals)

• The proposed quality index

7
Definition of The New Quality Index (con.)

• Where
• The dynamic range of Q is -1,1

8
Definition of The New Quality Index (con.)
(The original image signals)
(The change image signals)

• Ex.

9
Definition of The New Quality Index (con.)

• Rewrite the definition of Q

-1,1
0,1
0,1
Loss of correlation
Luminance distortion
contrast distortion
10
Definition of The New Quality Index (con.)

• Loss of correlation
• The best value 1 is obtained when
  for all
• i1,2,,N , where a and b are constants and a
  gt 0
• Luminance distortion
• It equals 1 if and only if
• contrast distortion
• The best value 1 is achieved if and only if

11
Application to Images
If there are a total of M steps , then the
overall quality index is given by
12
Application to Images (con.)
Original
Impulsive Salt-Pepper Noise
Additive Gaussian Noise
Multiplicative Speckle Noise
13
Application to Images (con.)
Mean Shift
Contrast Stretching
Blurring
JPEG Compression
14
Application to Images (con.)
15
Conclusion and Discussion

• Outperforms the MSE
• Simple
• Sensitive to the energy of errors