1. Tone mapping Digital Visual Effects, Spring 2007 Yung-Yu Chuang 2007/3/13 with slides by Fredo Durand, and Alexei Efros

2. Tone mapping How can we display it? Linear scaling?, thresholding? 10 -6 10 6 10 -6 10 6 Real world radiance Display intensity dynamic range Pixel value 0 to 255 CRT has 300:1 dynamic range

3. Preliminaries For color images Log domain is usually preferred. Gaussian filter. Sampling issues. Efficiency issues.

4. Eye is not a photometer! "Every light is a shade, compared to the higher lights, till you come to the sun; and every shade is a light, compared to the deeper shades, till you come to the night." — John Ruskin, 1879

5. We are more sensitive to contrast Weber’s law background intensity Just-noticeable Difference (JND) flash

6. Global operator (Reinhart et al) Approximation of scene’s key (how light or dark it is). Map to 18% of display range for average-key scene User-specified; high key or low key

7. low key (0.18)high key (0.5)

8. Frequency domain First proposed by Oppenheim in 1968! Under simplified assumptions, image = illuminance * reflectance low-frequency attenuate more high-frequency attenuate less

9. Oppenheim Taking the logarithm to form density image Perform FFT on the density image Apply frequency-dependent attenuation filter Perform inverse FFT Take exponential to form the final image

10. Fast Bilateral Filtering for the Display of High-Dynamic-Range Images Fr é do Durand & Julie Dorsey Laboratory for Computer Science Massachusetts Institute of Technology

11. A typical photo Sun is overexposed Foreground is underexposed

12. Gamma compression X  X  Colors are washed-out InputGamma

13. Gamma compression on intensity Colors are OK, but details (intensity high- frequency) are blurred Gamma on intensityIntensity Color

14. Chiu et al. 1993 Reduce contrast of low-frequencies Keep high frequencies Reduce low frequencyLow-freq. High-freq. Color

15. The halo nightmare For strong edges Because they contain high frequency Reduce low frequencyLow-freq. High-freq. Color

16. Durand and Dorsey Do not blur across edges Non-linear filtering OutputLarge-scale Detail Color

17. Edge-preserving filtering Blur, but not across edges Anisotropic diffusion [Perona & Malik 90] –Blurring as heat flow –LCIS [Tumblin & Turk] Bilateral filtering [Tomasi & Manduci, 98] Edge-preservingGaussian blurInput

18. Start with Gaussian filtering Here, input is a step function + noise outputinput

19. Start with Gaussian filtering Spatial Gaussian f outputinput

20. Start with Gaussian filtering Output is blurred outputinput

21. Gaussian filter as weighted average Weight of  depends on distance to x outputinput

22. The problem of edges Here, “ pollutes ” our estimate J(x) It is too different outputinput

23. Principle of Bilateral filtering [Tomasi and Manduchi 1998] Penalty g on the intensity difference outputinput

24. Bilateral filtering [Tomasi and Manduchi 1998] Spatial Gaussian f outputinput

25. Bilateral filtering [Tomasi and Manduchi 1998] Spatial Gaussian f Gaussian g on the intensity difference outputinput

26. outputinput Normalization factor [Tomasi and Manduchi 1998] k(x)=

27. outputinput Bilateral filtering is non-linear [Tomasi and Manduchi 1998] The weights are different for each output pixel

28. Contrast reduction Input HDR image Contrast too high!

29. Contrast reduction Color Input HDR image Intensity

30. Contrast reduction Color Intensity Large scale Fast Bilateral Filter Input HDR image

31. Contrast reduction Detail Color Intensity Large scale Fast Bilateral Filter Input HDR image

32. Contrast reduction Detail Color Intensity Large scale Fast Bilateral Filter Reduce contrast Large scale Input HDR image Scale in log domain

33. Contrast reduction Detail Color Intensity Large scale Fast Bilateral Filter Reduce contrast Detail Large scale Preserve! Input HDR image

34. Contrast reduction Detail Color Intensity Large scale Fast Bilateral Filter Reduce contrast Detail Large scale Color Preserve! Input HDR image Output

35. Oppenheimbilateral