«Publications of the University of Eastern Finland Dissertations in Forestry and Natural Sciences TUIJA JETSU Modeling color vision Publications of ...»
In publication [P2] (Cone Ratio in Color Vision Models), we considered how the changes of the cone ratio in the retina would affect color vision models. The basis of our analysis lies in a Multi-Stage Color Model by de Valois and de Valois (1993), and we show how changes in cone ratio affect the different stages of this model. The behavior of the Multi-Stage Color Model was tested with different cone ratios: 10:5:1, 18:5:1, 3:6:1, 12:1:1 and 1:1:1. The changes in cone ratio were implemented as changes in the weighting of the responses at the ﬁrst stage of the model. For the Farnsworth-Munsell 100 Hue color vision test, ratio 3:6:1 best preserved the organization of the test colors in the opponent color space. Ratios 12:1:1 and 1:1:1 differed most from the others. This article shows that if the changes in cone ratio would not be compensated for at all in the later stages of the human visual system, the resulting color space
would be very different for each individual.
In publication [P3] (Spectral images and the Retinex model), the color constancy performance of the Retinex algorithm in different color spaces was reviewed.The Retinex algorithm has earlier been applied mainly to grayscale or RGB images, but in this paper we consider different ways of applying the Retinex model to spectral images. The use of spectral images as the starting point for the model enables one, for example, more accurate examination of illumination or observer changes. First, we tested how the Retinex preserves colors when the spectral power distribution of the ambient light changes. To examine this, the algorithm was applied to images in four different color spaces: spectral, L channel of the L*a*b*, LMS, and sRGB. The color constancy between different illuminants was best preserved by using the spectral color space, and the poorest performance was reached by using the sRGB space.
When examining the behavior of the Retinex model in a case where the surround of an area of a scene changes, the results in RGB and spectral spaces are similar (contrary to the ﬁrst experiment) and closer to the real world color perception than the results in other color spaces. In general, it was noted that the output of the Retinex algorithm for all color spaces requires careful selection of the postprocessing method.
In publication [P4] (Color classiﬁcation using color vision models), we look at the color vision topic from a classiﬁcation point of view.
Instead of detecting and analyzing colors exactly in the same way, we have all just learned to classify colors in a certain way, which seems to lead almost always to the same result independent of the individual differences in the color vision system. The color classiﬁcation abilities of color vision models, considered also in [P1], were examined by using a simple subspace classiﬁcation method on Munsell Matte Collection color samples. The output of the MultiStage Color Model by De Valois and De Valois was a very poor starting point for color classiﬁcation, and with the output of Ingling and Tsou model the color classes were somewhat overlapping. The 41 Dissertations in Forestry and Natural Sciences No 20 Tuija Jetsu: Modeling Color Vision two nonlinear models, Bumbaca and Smith’s and Guth’s models, performed signiﬁcantly better in this classiﬁcation task. The errors made by the two latter models were similar to the ones human observers also often make, meaning that if the color was not classiﬁed into the correct class, it was usually classiﬁed into one of the neighboring classes.
In publication [P5] (The effect of stimulus color, size and duration in color naming reaction times), we examine the differences in color naming reaction times between subjects with normal and deﬁcient color vision. Color deﬁcient people use, in many cases, color names in a similar manner as people with normal color vision do. This means that the individual differences at the detection level of a color vision system are somehow compensated for at the later levels. In order to examine how different parameters affect the color naming process, we conducted a color naming experiment by using modiﬁed Berlin and Kay (1969) focal colors of different sizes and durations, and measured the reaction time for each stimulus.
Some differences were found between normal and color vision deﬁcient subjects. It was found that among the three examined parameters, color, size and duration, most differences in reaction times between color vision deﬁcient subjects and subjects with normal color vision were due to the color of the stimuli. Color vision deﬁcient (protan and deutan) subjects were clearly slower to react to red and green colors than to blue and yellow, whereas there were no remarkable differences in the reaction times for different colors with subjects having normal color vision. There was one deutan subject who consistently named red colors as green. In addition, subjects with normal color vision named very small blue stimuli as green. Also, the size of the stimuli had an effect on the reaction times for some subjects, and size and duration both affected the misclassiﬁcation rate of stimuli.
42 Dissertations in Forestry and Natural Sciences No 206 Conclusions As almost always when modeling natural phenomena, when modeling color vision the following questions also easily arise. Should the model be universal or speciﬁc just for a certain application? At what level of accuracy should the model be implemented? Is it even possible to create a universal model?
Understanding the behavior of human color vision requires knowledge from various ﬁelds of science. It makes no sense to claim that the overview presented here would be all-embracing, because there are countless possibilities for implementing the properties of human color vision. This dissertation and the related articles have been written with an intention to contemplate the properties of human color vision from a slightly different point of view.
It was found out in [P1] that none of the examined models were able to replicate the behavior of human color vision perfectly in all experiments. At least some kind of nonlinearity had to be implemented in order to be able to compensate for the differences between different brightness levels. Even though the performance of the Bumbaca and Smith model was the poorest with the tasks given in [P1], the model performed well in the color classiﬁcation task in [P4]. This is a good example of a model that performs well in a speciﬁc task for which it has been originally designed, in this case the performance being motivated by a computer vision application. Guth’s ATD model performed well in almost all given tasks, but from the ATD model it is not as easy to separate all the stages of the biological color vision system as it is, for example, in the De Valois and De Valois model. As models are based on different ideas, they also perform differently in given tasks.
Based on the results of [P2], we can speculate that if the only thing changing anatomically in the color vision system between individuals was the cone ratio on the retina, the personal color space of each individual would be quite different from others. Of course 43 Dissertations in Forestry and Natural Sciences No 20 Tuija Jetsu: Modeling Color Vision it is quite a radical simpliﬁcation of the process if only the cone ratio in the model is changed, and this does not tell the whole truth of what actually happens in the human color vision. When people have different cone distribution on the retina, the further neural connections related to color vision are most probably formulated in a different way, too. The actual spatial arrangement of the cones was not considered as a part of this dissertation, but it is also one interesting research topic related to color vision modeling.
Color constancy is one important property of the human color vision system, but it is not too easy to model, as was seen in [P3].
Also the connections between different colored stimuli and the actual color sensation can vary a lot depending on the parameters related to the stimuli, as was shown in the experiments in [P5]. The largest differences in color-related issues are typically seen between subjects with normal and deﬁcient color vision, but there are also individual differences within each group.
All these results strengthen the idea that color vision as a concept is complicated, and that the modeling of it well is a demanding task. The adaptive behavior of the color vision system is important from the modeling point of view: in order to be able to replicate the color vision accurately, a model must also have dynamic properties, like possibilities for spatial and temporal processing. It is very likely that no model can take into account every single aspect of color vision, but by combining the best properties of each model it is possible to ﬁnd a more universal approach.
44 Dissertations in Forestry and Natural Sciences No 20Bibliography  C. Balas, V. Papadakis, N. Papadakis, A. Papadakis, E. Vazgiouraki, and G. Themelis. A novel hyper-spectral imaging apparatus for the non-destructive analysis of objects of artistic and historic value. Journal of Cultural Heritage, 4(Supplement 1):330–337, 2003.
 D. Bimler and J. Kirkland. Colour-space distortion in women who are heterozygous for colour deﬁciency. Vision Research, 49:536–543, 2009.
 D. L. Bimler, J. Kirkland, and K. A. Jameson. Quantifying variations in personal color spaces: Are there sex differences in color vision? Color Research & Application, 29(2):128–134, 2004.
 J. Birch. Diagnosis of Defective Colour Vision, 2nd Ed.
Butterworth-Heinemann, Boston, 2001.
 F. Bumbaca and K. C. Smith. Design and implementation of a colour vision model for computer vision applications. Computer Vision, Graphics, and Image Processing, 39(2):226 – 245, 1987.
 J. Carroll, M. Neitz, H. Hofer, J. Neitz, and D. R. Williams.
Functional photoreceptor loss revealed with adaptive optics:
An alternate cause of color blindness. PNAS, 101(22):8461– 8466, 2004.
 V. Casagrande and T. Norton. Lateral geniculate nucleus: A review of its physiology and function. In J. Cronly-Dillon and A. G. Leventha, editors, Vision and visual dysfunction: The neural basis of visual function, volume 4, pages 41–84. Macmillan, London, 1991.
 E. J. Chichilnisky and B. A. Wandell. Trichromatic opponent color classiﬁcation. Vision Research, 39:3444–3458, 1999.
 T. Y. Chui, H. Song, and S. A. Burns. Adaptive-optics imaging of human cone photoreceptor distribution. J. Opt. Soc. Am. A, 25(12):3021–3029, 2008.
 T. Y. P. Chui, H. Song, and S. A. Burns. Individual variations in human cone photoreceptor packing density: variations with refractive error. Invest. Ophthalmol. Vis. Sci., 49(10):4679–4687, 2008.
 B. Cole. Assessment of inherited colour vision defects in clinical practice. Clinical and Experimental Optometry, 90:157–175., 2007.
 B. L. Cole. The handicap of abnormal colour vision. Clinical and Experimental Optometry, 87(4-5):258–275, 2004.
 B. R. Conway. Color vision, cones, and color-coding in the cortex. Neuroscientist, 15(3):274–290, 2009.
 B. R. Conway and M. S. Livingstone. Spatial and temporal properties of cone signals in alert macaque primary visual cortex. J. Neurosci., 26(42):10826–10846, 2006.
 H. D. Crane and T. P. Piantanida. On seeing reddish green and yellowish blue. Science, 221:1078–1080, 1983.
 B. de Gelder, M. Tamietto, G. van Boxtel, R. Goebel, A. Sahraie, J. van den Stock, B. M. Stienen, L. Weiskrantz, and A. Pegna.
Intact navigation skills after bilateral loss of striate cortex. Current Biology, 18(24):R1128–R1129, 2008.
 R. L. De Valois. Color vision mechanisms in the monkey. The Journal of General Physiology, 43(6):115–128, 1960.
 R. L. De Valois and K. K. De Valois. A multi-stage color model.
Vision Research, 33(8):1053–1065, 1993.
 S. S. Deeb and A. G. Motulsky. Molecular genetics of human color vision. Behavior Genetics, 26:195–207, 1996.
 P. DeMarco, J. Pokorny, and V. C. Smith. Full-spectrum cone sensitivity functions for X-chromosome-linked anomalous trichromats. J. Opt. Soc. Am. A, 9:1465–1476, 1992.
 A. M. Derrington, J. Krauskopf, and P. Lennie. Chromatic mechanisms in lateral geniculate nucleus of macaque. J Physiol., 357:241–265, 1984.
 M. Eng, P. Martin, and C. Bhagwandin. The analysis of metameric blue ﬁbers and their forensic signiﬁcance. Journal of Forensic Sciences, 54(4):841–845, 2009.
 D. Farnsworth. The Farnsworth-Munsell 100-hue and dichotomous tests for color vision. J. Opt. Soc. Am., 33(10):568–574, 1943.
 D. Farnsworth. The Farnsworth dichotomous test for color blindness panel D-15. Technical report, Psychological Corporation, New York, 1947.
 B. Funt, F. Ciurea, and J. McCann. Retinex in Matlab(TM). J.
Electron. Imaging, 13(1):48–57, 2004.
 P. Fält, J. Hiltunen, M. Hauta-Kasari, I. Sorri, V. Kalesnykiene, and H. Uusitalo. Extending diabetic retinopathy imaging from color to spectra. In Proceedings of the 16th Scandinavian Conference on Image Analysis, pages 149–158, Oslo, Norway, 2009.
 E. B. Goldstein. Sensation & Perception. Brooks/Cole, Paciﬁc Grove (CA), 5 edition, 1999.
 P. Gouras. Cortical mechanisms of colour vision. In P. Gouras, editor, Vision and visual dysfunction: The perception of colour, pages 179–197. Macmillan, London, 1991.
 S. L. Guth. Further applications of the ATD model for color vision. Proceedings of SPIE, 2414(1):12–26, 1995. Also in IS&T’s Recent Progress in Color Science, Karen Braun and Reiner
Eschbach, Eds. Society for Imaging Science and Technology, Springﬁeld, USA, 1997, pp. 177-185.
 S. L. Guth. The constancy myth, the vocabulary of color perception, and the ATD04 model. Proceedings of SPIE, 5292:1–14, 2004.
 T. Hansen, M. Giesel, and K. R. Gegenfurtner. Investigating human chromatic discrimination of natural objects. In CGIV 2006 – Third European Conference on Color in Graphics, Imaging and Vision, pages 119–124, Leeds, UK, 2006.