1. Validation of Color Managed 3D Appearance Acquisition Michael Goesele Max-Planck-Institut für Informatik (MPI Informatik) Vortrag im Rahmen des V 3 D 2 Workshops 29.+30. November 2004 in Berlin

2. Michael Goesele Acquired BRDF Model rendered BRDF model of a carafe Villeroy & Boch (Mettlach, 19 th century)

3. Michael Goesele Acquired BRDF Model Which model is (more) correct?

4. Michael Goesele Background: BRDF bidirectional reflectance distribution function (BRDF) ratio of reflected to incident radiance at one point for any pair of directions

5. Michael Goesele Background: BRDF Acquisition based on acquisition system for spatially varying BRDFs [Lensch et al. 2003] determine local reflection properties for each surface point uses Lafortune BRDF model [Lafortune 1997] shown at previous V 3 D 2 Workshops …

6. Michael Goesele Background: BRDF Acquisition

7. Michael Goesele Validation Questions How exact can this method capture describe the behavior of a real object? How exact can we reproduce an objects appearance? How can this be achieved?  will be discussed in this talk How good is the Lafortune model?  will not be discussed in this talk  see e.g. [Ngan et al., SIGGRAPH Sketch 2004]

8. Michael Goesele Color and Computer Graphics yes, we have color … it works (somehow) RGB is always the same (?) looks nice after some tuning … more seriously … not the main concern, other technologies more important eye can adjust to bad color reproduction comparison to ground truth often not possible/required an important issue for digitization!

9. Michael Goesele Background: Color Management goal: ensure correct color reproduction across devices used in graphical arts and printing industry defined profile connection space (PCS) with CIE XYZ or CIE Lab color space profiles describe conversion to PCS for all devices take limitations of devices into account

10. Michael Goesele ICC Profile Generation example: input profile capture known test target with camera lighting conditions (spectrum) identical to finally used conditions profile generated by (commercial) software captures properties of camera lighting test target

11. Michael Goesele Key Idea introduce color management into BRDF acquisition and reproduction workflow convert input images into defined color space (during HDR image generation) convert rendered images into output device color space allows for objective assessment of quality of acquired models important for libraries, conservation, … color management ensures best possible color reproduction takes limitations of devices into account

12. Michael Goesele BRDF Acquisition Workflow move to calibrated color space move to output color space

13. Michael Goesele Accuracy of the Acquired Model compare spectrophotometer measurements with BRDF model evaluated under same conditions illumination at 45°, observation at 0° (along surface normal) performs measurement in spectral domain can be converted to other color representations BRDFspectrophotometer

14. Michael Goesele Accuracy of the Acquired Model BRDF (CIEXYZ) Spectrophotometer (CIEXYZ) EE Bird yellow59.41, 61.23, 14.5861.40, 61.51, 13.424.158 Bird orange42.28, 33.17, 8.0243.73, 32.73, 5.957.923 Bird blue34.99, 40.09,46.6532.74, 37.48, 48.164.283 Bird white68.79, 73.35, 61.0476.67, 78.75, 58.777.168 Carafe blue17.63, 18.67, 33.2617.72, 19.34, 36.3253.398 Carafe white57.16, 58.33, 52.1154.00, 56.23, 43.277.003 quality metric:  E distance in CIELab color space 1  E just noticable color difference under perfect conditions

15. Michael Goesele Accuracy of the Acquired Model compare renderings with photographs captured under identical conditions quality metric:  E 0  E 50  E Minerva of Arezzo (Florence, 3rd century B.C. or 1st century A.C.)

16. Michael Goesele Accuracy of the Acquired Model until now: only comparison of acquired model to ground truth data important for conservation, long term storage further goal: include output devices in validation can we (physically) reproduce the object correctly?

17. Michael Goesele Accuracy of Renderings visual comparison between rendering (on screen, printout) and real object under identical conditions all devices are calibrated no manual color adjustment was performed! screenreal object (color laser printer) printout

18. Michael Goesele Accuracy of Renderings visual comparison between rendering (on screen, printout) and real object under identical conditions

19. Michael Goesele Accuracy of Renderings visual comparison between rendering (on screen, printout) and real object under identical conditions

20. Michael Goesele Conclusion color management integrated into BRDF acquisition and rendering pipeline enables quantitative and visual validation of results correct color acquisition and rendering is important! approach: acquire best possible model use best possible reproduction important for long term storage reproduction technology improves (displays, printers, …) model should support these as far as possible

21. Michael Goesele Future Work improve quality of color management some colors are still quite problematic color management for HDR images? ongoing work in the community handling of high contrast, out-of-gamut colors tone mapping problem new display technologies (e.g., HDR display)

22. Michael Goesele Thanks to … Hendrik Lensch DFG Schwerpunktprogramm V 3 D 2 “Verteilte Vermittlung und Verarbeitung digitaler Dokumente” More information … Michael Goesele, Hendrik P. A. Lensch, Hans-Peter Seidel: Validation of Color Managed 3D Appearance Acquisition. Proc. IS&T’s 12th Color Imaging Conference, pp. 265-270, 2004. Hendrik P. A. Lensch, Jan Kautz, Michael Goesele, Wolfgang Heidrich, Hans-Peter Seidel: Image-Based Reconstruction of Spatial Appearance and Geometric Detail. ACM Transactions on Graphics, vol. 22, 2, pp. 234-257, 2003.