RESEARCH NOTE WAVELENGTH DISCRIMINATION DERIVED FROM COLOR NAMING B. V. GRXXM’ School of Optometry and Department of Physiology and Biophysics. University of Alabama in Birmingham bl. E. TLXNER, JR. Department of Biostatistics. University of Alabama in Birmingham
R. HOLLAND Department
of Physiology and Biophysics, University of Alabama in Birmingham E. L. BRADLEY
Department of Biostatistics. University of Alabama in Birmingham and
J. A. BXJRDESHAW Southern Research Institute. Birmin~am.
It has been shown by Smith (1971) that the quantitative color naming method described by Boynton, Schafer and Neun (1964) and Boynton and Gordon (1965) can be used to investigate wavelength discrimination. In deriving wavelength discrimination from color naming scores, Smith (1971) did not make use of the variability in subject response to each waveIength across repeated observations; an important factor in specifying discriminating ability. Graham, Turner and Hurst (1973) have suggested an altemative method which takes this variability into account and which allows an estimation of wavelength discrimination magnitudes in units of information. This paper reports the application of their method to color naming data. Comparison with Wright’s (1947) wavelength discrimination results is made. The two methods are shown to be in essential agreement, confirming the usefulness of the color naming procedure. XfETHOD Apparatus The apparatus consisted of an illumination system and a perimeter screen which served as an adaptaiion background (Fig. I). In the illumination system a 1000 cc’xenon arc lamp was imaged onto the entrance slit of a hiah inten-
sity grating monochromator (GM) by a condeniing lens system L, and L2 and a cupric sulphate heat filter (HP). The grating monochromator delivered light of half-bandwidth 12 nm as controlled by the size of the entrance and exit slits. The light proceeded through lenses Ls, I..*, L, and L, to the fiber bundle. The light was attenuated by a neutral density filter (NDR and a ground glass plate (GG) t Present address: Ecole d’Optom&rie, Universite de Montreal, Case Postale 6128, Montreal, Quebec. Canada. jj9
which spatially homogenized the stimulus. An image of the exit slit of the monochromator fell on the Vincent shutter (S) which was controlled by digital circuitry to provide a 1 set pulse when the subject pressed a microswitch. The circular variable neutral density wedge (CVW) was rotated by means of a digital stepping motor and digital circuitry to yield equal luminances at the aperture (A). The equal luminances were set by the lP39 phototube (PT) onto which light was reflected by the microscope slide (MS) and which was calibrated against a Tektronix digital photometer placed at the aperture (A). The perimeter screen was that described by Graham (1972). It consisted of a half-cylindrical shell, whose diameter was 21 in. The subject’s left eye was positioned on the axis of the cylinder midway between the floor and the roof of the perimeter. The perimeter screen, which was spray-painted with Celofite flat white paint. was evenly illuminated with louvered 7.5 W Sylvania lamps run at 40 V, one above and one below the subject, to provide an adaptation level of approx @70 mL as measured by a Tektronix J&6 Digital photometer (corresponding approximately to 35 td as determined by employing the method described in Wyszecki and Stiles (19671, based upon the data of DeGroot and Gebhard (1932). We approximately equated the 550 nm stimuius in brightness to the background. and this luminance of the 550 nm stimulus was that used for all wavelengths and all subjects. The stimulus of dia 50’ was delivered via the fiber bundle and appeared in a I’ 24’ hole in the screen at the 0” perimetric location. The subject’s head and eye positions were controlled by means of an adjustable bite bar. Throughout the experiment the subjects left eye was used, and the right eye was occluded with an eye patch. Subjecrs.
The subjects were four volunteer college students with normal color vision as tested by the HRR. D-15 and Nagel Anomaloscope.
Research Note ILLUMlNATlON
Fig. 1. The apparatus consisted of an illumination system and a perimeter screen. linked by a fiber bundle (EB). The observer fixated the center of the aperture A with his left eye. The stimulus, which appeared at A, was spatially homogenized by the ground glass plate (GG). For a description of the illumination system see the text. Procrdm-r
At the beginning of each session an eye patch was placed
on the subject’s right eye. The subject sat with his head in the perimeter and light adapted for 10 min, after which he was positioned on the bite bar. He was instructed that for each stimulus presentation he would have to give a hue resoonse. consistina of 100% of Red. Green. Yellow or Blue or a percenta& mixtur; of pairs of these hues. The subject was instructed that yellow-blue and red-green responses were not allowed. The subject fixated the center of the aperture (A) and then initiated a trial by pressing a micro-switch. Each trial consisted of a I-set pulse. after which the subject gave his hue response. During each session two blocks of the wavelengths M-630 in IO-nm steps were presented. Each block was separately randomized in order of wavelength presentation. After a single training session, each subject was given five sessions. RESULTS
In order to catcuIate the derived wavelength discrimination, each response was coded into a nominal scale as follows: A response of looo/, of Red, Green, Yellow or Blue was coded as R, G, Y or B respectively. IMixed responses with a predominant hue were coded as RY, RB, GY, GB, YR, YG, BR. BG; the ‘This method extracts a common form, differing only in scale and origin. from the several sets of data; therefore, the four curves generated appear very similar. ‘Jacobs and Gaylord (1967) have demonstrated shifts in the position of unique yellow according to the spectral distribution of the adaptation light. However, comparison of their data with ours is difficult because of large differences in stimulus size and light level between the two experiments. The shift in unique yellow towards 600 nm is consistent with an adaptation light which was not perfectly spectrally neutral.
first letter indicating the predominant hue. Mixed responses in which both hues were judged to be 5P,, by the subject were coded as R’Y. R2B, G’Y and G’B. Since no mixed responses of red and blue were elicited, there were 13 response types. The ten observations at each wavelength generated a frequency of each response type at each wavelength. In order to test whether a pair of wavelengths w-as di~~~at~ we tested for the null hypothesis that the sample frequency distributions of response types for each wavelength were drawn from the same population frequency distribution of response types. This analysis also allowed a measure of the magnitude of discriminatory power and we calculated a measure of discrimination information, in bits. The details of this statistical method have been previously described (Graham, Turner and Hurst, 1973). The weighted mean discrimination magnitude was calculated hetween each wavelength and its two neighboring wavelengths. The results, in units of information (bits) are plotted in Fig. 2. Each subject’s calculated wavelength discrimination data show two maxima. These data. when graduated by “self-modeRing nonlinear regression”2 (Lawton et nf., 1972), suggest the possibility of an inflexion in the middle region of the spectrum. More subjects will be needed before the existence of this inhexion can be confirmed or disproved. The color naming data are sho&n in Fig. 3. They are plotted in per cent point values which are the mean per cent responses for each of the hues (Red. Green, Yellow and Blue) at each wavelength. Each color naming graph shows a crossover from red responses to green responses close to 600 nm.3 The position of the crossover of yellow to blue is less reliably placed in the four subjects. The steepness of crossin_n of the red and green curves is well-marked in all four
600W 65OW WAVE LENGTH hd
6oo.w 65000 WAVE LENGTti hml
Fig. 2. The derived wavelength discrimination data of the four subjects (DM, TC, TF and GJ) are shown. Each subject had 10 observations at each wavelength. The curves were graduated by “self-modeling nonlinear regression” (Lawton er al., 1972). Each set of data shows two maxima. There is a suggestion of an inflexion in each data set around the middle of the spectrum. The similarities of magnitudes of the curves of the different subjects are notable.
Loo 475 v2a
600 62) 650 LENGTH (nml
95 WAVE LENGTH
Fig. 3. The color naming responses are shown for the four subjects. The per cent point values are the mean per cent responses in each of the hues red, green, yellow and blue. The zero response for any hue occurs somewhere between the zero response on the graph and the closest non-zero response. The crossing of the red and green curves occurs in all four subjects close to 600 nm. The yellow-blue crossing is quiet variable across the four subjects. The blue-green crossing occurs close to 500 nm in all four subjects.
subjects as compared to the shallouer crossing of the yellon and blue response curves. a feature especial11 marked in subjects TF and GJ. In all four subjects the intersection of the blue and green rssponse cur\es is steep and its position is reliably close to 500 nm. DISXSSIOS The wavelength discrimination curl-es obtained demonstrate the application of the statistical method previously proposed to derive wavelength discrimination from color naming (Graham. Turner and Hurst. 1973). Furthermore. it is notable that the information peaks in the curves are very close in magnitude across the four subjects. By virtue of the additivity property of information. the number of bits of information at these peaks is dependent partially on the number of presentations of each observation (see for example. Khinchin, 1957). However. each subject uas presented with the same number of observations at all wavelengths. The reliability in magnitude of bvavelength discrimination suggests that with a large enough sample of subjects useful norms may be obtained. We now turn to a comparison of the color naming data with the derived kvavelength discrimination curves. One maximum in the wavelength discrimination curve coincides with the intersection of the red and green response curves: :he other maximum with the intersection of the blue and green response curves rather than the intersection of the yellow and blue response curves. Wright (1947), from a bipartite field study performed at approximately 70 “photons” (i.e. 70 td). observed that maximum discrimination occurs in the spectrum where there is a change from reddish-yellow to greenish-yellow. from greenish-blue to bluish-green and in the violet, where there is a change from reddish-blue to bluish-red. Since we did not examine wavelengths below 450 nm, no statement can be made here about the maximum in the violet
oi the spectrum.
results JeriLed :‘rom c~ii?r n.i:i‘-
ing are in essential +reement with :hose of L\‘r;f!:r by direct compariwn of ii:<, :\A,? hnl\,es of a bipartite tklii.
-Ic,b,lo~~le[iyrrtl~tfrs--Thes.xperimentai portion of [hi> uori was performed in the Neurosciences Progam. The Clzdicai Center. University of .Alabama in Birmingham.
Boynton R. M.. Schafer \V. and Nsun \I. -\. (196-Ij Hue wavelength relation measured by color-naming method for three retinal locations. Science la. 666-665. Boynton R. M. and Gordon J. 11964) Bezold-Brucks hue shift measured by color-naming technique. J. qlr. Sot. .4m. 55,
DeGroot S. G. and Gcbhard J. W. (1952) Pupil size as determined by adapting luminance. J. c~pr. Sot. .4m 41. -192. Graham B. 1’. 11972 Color Lision in the peripheral iisual field. Ph.D. Dissertation. Indiana L’niv. Available through Xerox Microfilms. Ann Arbor. Mich. Graham B. V.. Turner M. E. and Hurst D. C. (19Yj1 Derivation of wavelength discrimination irom color naming J. opf. Sot. ,4m. 63. (No 1I. 109-l II. Jacobs G. H. and Gaylord H. A. 1196-1 Effects of shromatic adaptation on color naming. L’isiorr Rrs. 7. 645-65-l.
Khinchin .A.I. (1957) Marherrwriccll Fo~~rA~rionsof‘fqtirm~~tion Theor.~. Dover. New York. Lawton W. H.. Sylvestre E. A. and Slzzgio Xl. F. (19’31 Self modeling nonlinear regression. T~chr10niefr1c.s14. (No. 3). 513-532. Smith D. P. (1971) Derivation of ucl;slength discrimination from colour-naming data. l’ision Rrx. I I, ?39-741, U’right W. D. (1947) Researchrs on .Vormd cd D&c~tice Co/our L’ision. pp. 167-168. Mosby. St. Louis. K’yszecki G. and Stiles W. S. I 19671 C&r Scierrcr. pp. 213-21-1. Wiley. New York.