Introduction to Image Quality Assessment

1
Introduction to Image Quality Assessment
2
Outline

• Applications
• Image Quality Assessment
• Image Quality Assessment with Reference Image
• -Zhou Wang and Alan C. Bovik, ICASSP2002
• Blind Image Quality Assessment
• -Xin Li, ICIP2002

3
Image Quality ?
4
Image Quality Assessment
Good
Bad
5
Applications

• Image Acquisition Systems and Display Systems
• Image Processing Systems and Algorithms
• Compression and Network

6
Image Quality Assessment

• Mean Opinion Score
• Automatically Image Quality Evaluation

7
Mean-Squared Error and Peak Signal-to-Noise Ratio
8
Frequency-domain SNR
9
Frequency-domain SNR
10
Image Quality Assessment Methods

• Image Quality Assessment with Reference Image
• Blind Image Quality Assessment

11
Image Quality Assessment Methods

• Image Quality Assessment with Reference Image
• Blind Image Quality Assessment

12
Error Sensitivity Based Image Quality Measurement
13
Error Sensitivity Based Image Quality Measurement
alignment, luminance transformation, and color
transformation
14
Error Sensitivity Based Image Quality Measurement
resulting in two sets of transformed signals for
different channels
15
Error Sensitivity Based Image Quality Measurement
The errors between the two signals in each
channel are calculated and weighted, usually by a
Contrast Sensitivity Function (CSF).
16
Error Sensitivity Based Image Quality Measurement
The weighted error signals are adjusted by a
visual masking effect model, which reflects the
reduced visibility of errors presented on the
background signal
17
Visual Masking Effect
18
Error Sensitivity Based Image Quality Measurement
Minkowski error pooling
19
Weaknesses of Error Sensitivity Based Methods

• The reference signal is of perfect quality
• There exist visual channels in the HVS and the
  channel responses can be simulated by an
  appropriate set of channel transformations.
• CSF variance and intra-channel masking effects
  are the dominant factors that affect the HVSs
  perception on each transformed coefficient in
  each channel

OK!
20
Weaknesses of Error Sensitivity Based Methods

• For a single coefficient in each channel, after
  CSF weighting and masking, the relationship
  between the magnitude of the error and the
  distortion perceived by the HVS can be modeled as
  a non-linear function.
• The interaction between channels is small enough
  to be ignored.

21
Weaknesses of Error Sensitivity Based Methods

• The perceived image quality is determined in the
  early vision system. Higher level processes, such
  as feature extraction, pattern matching and
  cognitive understanding happening in the human
  brain, are less effective
• Active visual processes, such as the change of
  fixation points and the adaptive adjustment of
  spatial resolution because of attention, are less
  effective

22
Structure Distortion Based ImageQuality
Measurement

• The main function of the human eyes is to extract
  structural information from the viewing field,
  and the human visual system is highly adapted for
  this purpose. Therefore, a measurement of
  structural distortion should be a good
  approximation of perceived image distortion

23
Structure Distortion Based ImageQuality
Measurement
24
Structure Distortion Based ImageQuality
Measurement
loss of correlation, mean distortion, and
variance distortion
25
Experimental Results
a original b 0.9372 c 0.3891 d 0.6494 e
0.3461 f 0.2876
26
Image Quality Assessment Methods

• Image Quality Assessment with Reference Image
• Blind Image Quality Assessment

27
Blind Image Quality Assessment

• Human visual system usually does not need any
  reference to determine the subjective quality of
  a target image
• Distinction between fidelity and quality

28
Blind Image Quality Assessment

• Edge Sharpness Level
• Random Noise Level
• Structured Noise Level

29
Edge Sharpness Level
30
Random Noise Level

• Impulse Noise
• Additive White Gaussian Noise

31
Structured Noise Level

• Block Artifact
• Ringing Artifact

32
Blind Image Quality Assessment

• Combination of different measurement is still a
  problem.

33
Conclusion

• Image Quality Assessment with Reference Image
• Blind Image Quality Assessment