Dynamic Range Independent Image Quality Assessment Tun Aydin*, Rafa - - PowerPoint PPT Presentation

▶

Jan 06, 2024 171 likes •646 views

Dynamic Range Independent Image Quality Assessment Tun Aydin*, Rafa Mantiuk, Karol Myszkowski and Hans-Peter Seidel MPI Informatik Image Quality Assessment Example Applications Image Compression Global Illumination Image Processing

SLIDE 1

Dynamic Range Independent Image Quality Assessment

Tunç Aydin*, Rafał Mantiuk, Karol Myszkowski and Hans-Peter Seidel MPI Informatik

SLIDE 2

Image Quality Assessment

SLIDE 3

Example Applications

Global Illumination

Speed up rendering without affecting quality [Myszkowski 2002] Benchmarking systems and algorithms [Dabov 2008]

Image Processing Image Compression

How much compression without visible artifacts?

SLIDE 4

Subjective Experiments

Quality of the distorted image ?

Rate the Quality

+ Reliable - High cost

SLIDE 5

Simple Quality Metrics

MSE ~ 280 MSE ~ 280 MSE ~ 280 !

Based on the Differences between Images

Examples: Mean Squared Error (MSE), Peak Signal to Noise Ratio (PSNR) Random Noise Blur ~15% Decreased Luminance Reference

SLIDE 6

Human Visual System (HVS) Based Metrics

Based on the Visible Differences between Images

Examples: Visible Differences Predictor (VDP) [Daly 93], HDR-VDP [Mantiuk et al. 05], Visual Discrimation Model (VDM) [Lubin 95]

Probability of Detection: ~15% Decreased Luminance Random Noise Distortion Map Distortion Map

SLIDE 7

From Distortion Magnitude to Structural Similarity

Measure the Preservance of Image Structure

Examples: Structural Similarity Index Metric (SSIM) [Wang et al. 04] Reference Contrast Enhancement

Distortions may be

Visible

but

Not

bjectionable

SLIDE 8

Visibility Structure Simple metrics No No HVS based metrics Limited or full dynamic range No Structural similarity based metrics Calibration challenging due to “abstract” parameters Yes Our approach: Hybrid of HVS and Structural Similarity

SLIDE 9

Focus Point: Image Pair with Different Dynamic Ranges

Similar appearance … … yet very different luminance

SLIDE 10

Outline

Detecting visibility thresholds
Full Dynamic Range Human Visual System

(HVS) Model

Detecting structural changes
A set of new distortion measures
Advantages over previous work
Possible applications

SLIDE 11

Human Visual System (HVS) Model

…of the entire visible dynamic range

SLIDE 12

Human Visual System Model

Luminance Masking Contrast Sensitivity Light Scattering

[ JND ] [ LUMINANCE ]

Channel Decomposition

SLIDE 13

Luminance Masking Contrast Sensitivity Light Scattering

Channel Decomposition

Decreased sensitivity due to glare around bright spots [Deeley et al. 1991]

SLIDE 14

Luminance Masking Contrast Sensitivity Light Scattering Decreased sensitivity due to glare around bright spots [Deeley et al. 1991]

Channel Decomposition

SLIDE 15

Luminance Masking Contrast Sensitivity

Channel Decomposition

Light Scattering Log Luminance # of JNDs Transform image luminance to Just Noticeable Difference (JND) Space [Mantiuk et al. 2005]

SLIDE 16

Luminance Masking Contrast Sensitivity Light Scattering

Channel Decomposition

Low Sensitivity Low Sensitivity

Decreased Sensitivity of very low and high frequencies [Daly 1993] Spatial Freq. Contrast

SLIDE 17

Luminance Masking Contrast Sensitivity

Channel Decomposition

Light Scattering

… …

. . . . . .

6 Frequency Bands 6 Orientations Low Pass Image Cortex Transform [Watson 1987, Daly 1993]

SLIDE 18

Distortion Measures

SLIDE 19

Loss of Visible Contrast

REFERENCE Contrast Visibility Threshold

SLIDE 20

Loss of Visible Contrast

TEST REFERENCE Contrast Visibility Threshold

SLIDE 21

Loss of Visible Contrast

Reference Test (Clipping) Distortion map

SLIDE 22

Amplification of Invisible Contrast

REFERENCE Contrast Visibility Threshold

SLIDE 23

Amplification of Invisible Contrast

TEST REFERENCE Contrast Visibility Threshold

SLIDE 24

Amplification of Invisible Contrast

Reference Distortion map* Test (Contouring) *For clarity, visible contrast loss is not shown

SLIDE 25

Reversal of Visible Contrast

Contrast REFERENCE

SLIDE 26

Reversal of Visible Contrast

Contrast Visibility Threshold Visibility Threshold TEST REFERENCE

SLIDE 27

Reversal of Visible Contrast

Reference Local contrast reversal

SLIDE 28

No Structural Distortion

Visibility Threshold Visibility Threshold

SLIDE 29

Visualization

SLIDE 30

Advantages over previous metrics

SLIDE 31

Case Study

Local Gaussian Blur HDR Test HDR Reference LDR Test LDR Reference

SLIDE 32

(1) HDR pair

HDR-VDP Our Metric SSIM

Loss Amplification Reversal

Distortion

SLIDE 33

(2) LDR pair

HDR-VDP Our Metric SSIM

Loss Amplification Reversal

Distortion

SLIDE 34

(3) HDR test, LDR reference

HDR-VDP Our Metric SSIM Distortion

Loss Amplification Reversal

SLIDE 35

(4) LDR test, HDR reference

HDR-VDP Our Metric SSIM

Loss Amplification Reversal

Distortion

SLIDE 36

Detecting distortions

HDR-VDP SSIM Sharpening Blur REFERENCE

SLIDE 37

Detecting “types” of distortions

Our Method Sharpening Blur REFERENCE

Loss Amplification Reversal

SLIDE 38

Applications

SLIDE 39

TMO Evaluation

FATTAL PATTANAIK

Loss Amplification Reversal

REFERENCE

SLIDE 40

Inverse TMO Evaluation

Loss Amplification Reversal

REFERENCE LDR2HDR

SLIDE 41

Display Comparison (1)

BrightSide DR37-P HDR Display (2000 cd/m2)

REFERENCE

Loss Amplification Reversal

SLIDE 42

Loss Amplification Reversal

Display Comparison (2)

Barco Coronis 3MP LCD Display (400 cd/m2)

SLIDE 43

Loss Amplification Reversal

Display Comparison (3)

Samsung SGH-D500 Cell Phone Display (30 cd/m2)

SLIDE 44

Summary

Hybrid approach: HVS and structure
Comparing different dynamic ranges
Detecting “type” of distortions
Applications on (inverse) tone mapping

and display comparison

TODO

– Color – Supra-Threshold

SLIDE 45

SLIDE 46

SLIDE 47

Luminance Masking Contrast Sensitivity Light Scattering Decreased sensitivity due to glare around bright spots [Deeley et al. 1991]

Channel Decomposition