Apparent Greyscale: A Simple and Fast Conversion to Perceptually Accurate Images and Video

1. Apparent Greyscale: A Simple and Fast Conversion to Perceptually Accurate Images and Video Kaleigh Smith Pierre-Edouard Landes Joelle Thollot Karol Myszkowski

2. OUTLINE • Introduction • Related Work • Apparent Lightness • Global Apparent Lightness Mapping • Local Chromatic Contrast Adjustment • Result • Conclusion

3. Introduction • We use a two-step approach for converting complex images and video to perceptually accurate greyscale versions. • 1. globally assign grey values and determine color ordering. • 2. locally enhance the greyscale to reproduce the original contrast

4. Introduction • Our global mapping is image independent and incorporates the Helmholtz-Kohlrausch effect for predicting differences between isoluminant colors. • We are not too sensitive to the loss of discriminability when it occurs between spatially distant colors, but with adjacent colors it is immediately apparent.

5. Related Work • [Gooch et al.]find grey values that best match the original color differences through an objective function minimization process. • [Rasche et al.]propose a similar approach that finds the linear transform matching pairwise grey differences to corresponding color differences. • [Neumann et al.]present a technique with linear complexity that requires no user intervention.

6. Apparent Lightness • Throughout this paper, we work in the CIELAB and CIELUV color spaces, whose three axes approximate perceived lightness, saturation and hue angle. • The first component, L*, quantifies the perceptual response of a human viewer to luminance and is defined as L* = 116(Y/Y0)1/3 −16 for luminance Y and reference white luminance Y0.

7. Apparent Lightness • While luminance is the dominant contributor to lightness perception, the chromatic component also contributes, and this contribution varies according to both hue and saturation. • The phenomenon is characterized by the Helmholtz-Kohlrausch effect, where given two isoluminant colors, the more colorful sample appears brighter.

8. Apparent Lightness • three predictors to correct L* based on the color’s chromatic component.

9. Apparent Lightness • We now decide which predictor is best suited to greyscale conversion. • In testing L*VCC , we observe that its stronger effect maps many bright colors to white, making it impossible to distinguish between very bright isoluminant colors. • L** exhibits a small range at blue hues. This range reduction makes L** becomes less discriminable. • We therefore conclude that L*VAC is the most suitable H-K predictor to use.

10. Global Apparent Lightness Mapping • The mapping process is as follows: • We first convert the color image to linear RGB by inverse gamma mapping, then transform to CIELUV color space. • Its apparent chromatic object lightness channel L*VAC is calculated according to (2). We map L*VAC to greyscale Y values using reference white chromatic values for u* and v*. • Finally, we apply gamma mapping to move from linear Y space back to a gamma-corrected greyscale image G

11. Global Apparent Lightness Mapping • Due to the compression of a 3D gamut to 1D, L*VAC may map two different colors to a similar lightness, which then are quantized to the same grey value. • This occurs only when colors differ uniquely by hue, which is very uncommon in natural images and well-designed graphics. • Our global mapping partially solves the problem of grey value assignment and appropriately orders colors that normal luminance mapping can not discriminate.

12. Local Chromatic Contrast Adjustment • Because of dimension reduction and unaccounted for hue differences, chromatic contrast may be reduced. • Humans are most sensitive to these losses at local contrasts, regions where there is a visible discontinuity. • To counter the reduction, we increase local contrast in the greyscale image G to better represent the local contrast of original I.

13. Local Chromatic Contrast Adjustment • We perform contrast adjustments using the Laplacian pyramid that decomposes an image into n bandpass images hi and a single lowpass image l • At each scale in the Laplacian pyramid, we adaptively increase local contrast hi(GL*) by a perceptually-based amount λi, which measures the amount of contrast needed to match color contrast hi(I).

14. Local Chromatic Contrast Adjustment

15. Result

16. Result

17. Result

18. Result

19. Conclusion • We have presented a new approach to color to grey conversion. Our approach offers a more perceptually accurate appearance. • The main limitation of our approach is the locality of the second step (local contrast adjustment). It can not restore chromatic contrast between non-adjacent regions. • This step also risks introducing temporal inconsistencies.