A Model of Local Adaptation

Transcription

1 A Model of Local Adaptation Peter Vangorp Karol Myszkowski Erich W. Graf Rafał K. Mantiuk Bangor University and MPI Informatik MPI Informatik University of Southampton Bangor University and University of Cambridge

2 Contrast Perception Would you be able to read the print on the light bulb? 2

3 Contrast Perception 3

4 Contrast Perception 4

5 Previous Models of Adaptation Physiology and psychophysics Naka-Rushton: L a requires adaptation luminance L a Ad hoc models used in computer graphics Naka-Rushton model with adaptation luminance =? global average luminance local per-pixel luminance local average computed in 1 Gaussian window 5

6 Basic Contrast Detection Model input luminance image L OTF tvi / traditional threshold contrast ΔL/L 6

7 Our Adaptation Model input luminance image L OTF tvi traditional threshold contrast ΔL/L / Local Adaptation adaptation luminance L a elevated threshold contrast ΔL/L Threshold Elevation 7

8 Our Adaptation Model input luminance image L OTF tvi Local Adaptation: traditional threshold contrast ΔL/L NL Pooling NL -1 / adaptation luminance L a elevated threshold contrast ΔL/L Threshold Elevation 8

9 Our Adaptation Model input luminance image L OTF tvi Local Adaptation: traditional threshold contrast ΔL/L NL 1 Pooling NL NL 2 Pooling NL / + adaptation luminance L a elevated threshold contrast ΔL/L Threshold Elevation 9

10 Our Adaptation Model input luminance image L OTF tvi Local Adaptation: traditional threshold contrast ΔL/L NL 1 Pooling NL NL 2 Pooling NL / + adaptation luminance L a elevated threshold contrast ΔL/L Threshold Elevation 10

11 High Dynamic Range Display 11

12 High Dynamic Range Display ipad LCD (retina resolution) 12

13 High Dynamic Range Display Projector backlight cd/m 2 13

14 Experiments: Baseline Adaptation 1. Adapt similar to [Hood et al. 1979] 14

15 Experiments: Baseline Adaptation 2. Flash 200 ms similar to [Hood et al. 1979] 15

16 Experiments: Baseline Adaptation 3. Orientation of the edge? similar to [Hood et al. 1979] 16

17 Experiments: Baseline Adaptation 3. Orientation of the edge? or? similar to [Hood et al. 1979] 17

18 Experiments: Baseline Adaptation Adaptation luminance [cd/m 2 ] 18

19 Experiments: Baseline Adaptation Adaptation luminance [cd/m 2 ] 19

20 Experiments: Baseline Adaptation contrast vs. intensity Adaptation luminance [cd/m 2 ] 20

21 Experiments: Baseline Adaptation contrast vs. intensity Adaptation luminance [cd/m 2 ] 21

22 Experiments: Baseline Adaptation contrast vs. intensity Adaptation luminance [cd/m 2 ] 22

23 Experiments: Adaptation Patterns Extent of pooling Long-range effects Pooling non-linearity Radial symmetry & contrast masking Natural images 23

24 Experiments: Extent of pooling 1. Adapt 24

25 Experiments: Extent of pooling 2. Flash 200 ms 25

26 Experiments: Extent of pooling 3. Orientation of the edge? 26

27 Experiments: Extent of pooling Detection contrast levels off at

28 Experiments: Pooling non-linearity Not linear Not logarithmic Asymmetric central flash luminance Bright half luminance [cd/m 2 ] 28

29 Experiments: Radial symmetry Rotation makes no difference 29

30 Experiments: Adaptation Patterns Extent of pooling Long-range effects Pooling non-linearity Radial symmetry & contrast masking Natural images 30

31 Model Fitting and Cross-validation General model 56 specific candidate models Model fitting using parallel genetic optimization Cross-validation: maximally differentiating stimuli 31

32 Our Best Adaptation Model Wider support at lower luminance due to non-linearities adaptation site shifts to postreceptoral mechanisms [Dunn et al. 2007] Complex pooling mechanism cross-validated to avoid overfitting more complex than known retinal pooling receptive fields in LGN or visual cortex? 32

33 Our Best Adaptation Model Wider support at lower luminance due to non-linearities adaptation site shifts to postreceptoral mechanisms [Dunn et al. 2007] Complex pooling mechanism cross-validated to avoid overfitting more complex than known retinal pooling receptive fields in LGN or visual cortex? 33

34 Our Best Adaptation Model Wider support at lower luminance due to non-linearities adaptation site shifts to postreceptoral mechanisms [Dunn et al. 2007] Complex pooling mechanism cross-validated to avoid overfitting more complex than known retinal pooling receptive fields in LGN or visual cortex? 34

35 Our Best Adaptation Model Wider support at lower luminance due to non-linearities adaptation site shifts to postreceptoral mechanisms [Dunn et al. 2007] Complex pooling mechanism cross-validated to avoid overfitting more complex than known retinal pooling receptive fields in LGN or visual cortex? 35

36 Our Best Adaptation Model Wider support at lower luminance due to non-linearities adaptation site shifts to postreceptoral mechanisms [Dunn et al. 2007] Complex pooling mechanism cross-validated to avoid overfitting more complex than known retinal pooling receptive fields in LGN or visual cortex? 36

37 Image Sample density Application: Adaptive Rendering Adaptive sampling until noise contrast undetectable Our method Non-adaptive equal time Weber adaptive equal quality 30% slower 37

38 Application: HDR Display Design Front LCD panel Backlight LED array Combined HDR image Visible difference

39 Application: HDR Display Design Front LCD panel Backlight LED array Combined HDR image Visible difference

40 Application: HDR Display Design Front LCD panel Backlight LED array Combined HDR image Visible difference

41 Application: Dynamic Range 41

42 Application: Dynamic Range 42

43 Application: Afterimages 43

44 Application: Gaze-dependent TM 44

45 Summary of Contributions General model local adaptation luminance contrast detection threshold Experiment contrast detection while adapted to various patterns Analysis interpretation of results of individual sets of patterns model fitting to all patterns cross-validation using maximally differentiating patterns A selection of applications 45

46 Thanks! Questions? Source code available at Acknowledgments: High Performance Computing Wales Fraunhofer and Max Planck cooperation within German Pact for Research and Innovation (PFI)