Perpmon Press Ltd 1’419. Rtntcd an Great Brttrm

UNIQUE ARE NOT INVARIANT WITH BRIEF STIMULUS DURATIONS’

Arriz~ L. NAGY’ University of Michigan

(Received I8 Seprrmber 197s)

Abstract-Unique loci were obtained from three observers at several intensity levels with a I-set stimulus flash and a 17 msec stimulus flash. With the one-second stimulus the three spectral unique hues were approximately invariant but unique was not. With the I7 msec stimulus none of the unique hues is strictly invariant. These results are discussed in terms of their implications for the nature of the cone signal inputs to the opponent mechanisms.

IL\iTRODUCTION recently been provided by Krantz (1975) and Larimer et al. (1974, 19X), whose work revolves around the The relationship between uniques hues and invariant question of whether unique hues are invariant hues hues has long been of interest in the study of color and whether additions of hues unique with respect to perception. Unique hues are the “psychologically” the same mechanism (e.g. unique and unique primary or “pure” hues, which do not appear to be , which both are unique with respect to the red/ compounds of other hues: red, , yellow. and green mechanism) result in another unique hue. These blue. The term invariant hues refers to those wave- two properties follow from the assumptions of fengths or spectral composites whose perceived hue modem opponent theory (Jameson and Hur- does not vary with changes in stimulus intensity. It is vich, 1955; Krantz, 1975), which include: clear that the hue of most stimuli does vary with stimulus intensity (Purdy, 1931). The effect is com- (i) Linear recombination of signals initiated by pho- monly known as the Bezold-Briicke effect. topigment absorptions in cones into two opponent As early as 1880 Hering maintained that the set of color mechanisms; red/green and yellow/blue. unique hues was identical with the set of invariant (ii) Faster growth of yellow/blue opponent mechan- hues. This identity has significance for models of the ism signals than red/green signals as intensity is in- color coding mechanisms and explanations of the in- creased. tensity dependence of most hues. The intensity depen- (iii) Dependence of perceived hue on the ratio of the dence of hue has been considered to be a manifes- two opponent mechanisms responses and in particu- tation of non-linearity in the color coding mechanism lar, observance of unique hues when the response of nearly since the time of its discovery. This idea is one opponent mechanism is zero. generally credited to Pierce (1887). though some Together these assumptions predict that all unique credit should be given to Maxweil (1856) for first hues must be invariant hues and conversely that all recognizing the non-linearity of color coding. How- invariant hues must be unique hues. Additions of two ever, there has been some debate about where the hues unique with respect to the same mechanism non-linearity occurs. Trichromatic theories of color must produce another unique hue. All hues which are coding have suggested that the hue shifts occur as a not unique must vary with stimulus intensity as a result of non-linearities in the signals from the three result of the faster growth of yellow/blue opponent cone mechanisms (Pierce, 1877; Walraven, 1961; mechanism signals. Savoie, 1973). Opponent theories of color coding have From their experiments Larimer et al. conclude suggested that the signals input to the opponent that assumption (i) is correct for the red/green mech- mechanisms from the three cone mechanisms are anism but not for the yellow/blue mechanism They linear with intensity, and that the non-linearity which found that unique red was not invariant, which im- produces the hue shifts occurs in the signals of the plies that signals input to the yellow/blue mechanism opponent mechanisms (Judd, 1951; Jameson and are not linear functions of intensity. They point out, Hurvich, 1955). Some resolution to this debate has however, that the departure from linearity would tend to work against the Bezold-Briicke hue shifts. This finding implies that the Bezold-Briicke hue shift is ’ This research was supported by NE1 Post Doctoral primarily a manifestation of non-linearity more proxi- Fellowship 7 F32 EY-05083 to the author, and NSF grant mai &an the site of opponent cancehation, as sug- 33642-X to David H. Krantz. Portions of this paper were gested by assumption to (ii). Though there is some presented at the Spring Meetings of the Association for Research in Vision and Opthalmology. Sarasota. Florida non-linearity in input to the yellow/blue mechanisms, 1978. its effect on perceived hue must be small compared to ’ Current address: Center for Human Information Pro- the effect of the non-linearity at the opponent level cessing, University of California. San Diego, California which primarily determines how the hue of most 92093, U.S.A. stimuli will change with intensity.

“It. 19 I?-_1 1427 Recent experiments (Savoie. 1973; Nagy and Zacks. to the rules of J double random st;llrcase proccdurs 1977: Center et a/., 197s) have suggested that unique \Cornsueet 1962). The starting points of the two rsndomlq yellow is not invariant with intensity when the stimu- intermixed staircases. generali) 30 or -lOnm apart. mere lus is a brief flash. This result implies that assumption chosen to bracket the region containing the unique hue. (i) may not hold for the red:green mechanism as well Each staircase independentl! followed the same rules for step sizes. The initial step %ze was 10 nm. After the first as the yellow blue mechanism under some conditions. reversal, the step size was 5 nm until the second reversal It also suggests that the degree of linearity or non- occurred. After the second reversal. the step size was fixed linearity of cone signals input to the opponent mech- at either two or three nm. The staricase was terminated anisms may be dependent to some degree on the par- \bhen three reversals had occurred with the smallest step ticular stimulus conditions used. This paper presents size. After both staircases were completed the luminance of further evidence for this suggestion. Results show that the stimulus was changed and another pair of staircases the invariance of unique hues is clearly dependent on igas run. Luminance was barisd in 0.55 log unit steps over stimulus duration. implying that the nature of the the available range generally going from bright to dim or intensity-response functions of the signals input to the dim to bright. Trials were presented ebery I@-20s~ with opponent mechanisms is also dependent on stimulus brief rest periods between pairs of staircases. In order to obtain a unique red. 625 and -170nm lights were mixed. duration. The 625 nm light was yellou-red in hue and the -170 nm for most observers. .4ddition METHODS ligh! uas nearI! unique blue of small amounts of 170 nm light to the 615 nm light can- .A conventional Maxwellian view optical system was celled out the yellowness in the 625 nm light resulting in ;L used to deliver stimuli to the observer. A Xenon-arc lamp unique red. which was quite saturated in chromatic (150 W Osram) driven by a current-regulated d.c. power appearance. Slightly more saturated could have been supply provided the illumination which was passed produced by adding the J7Onm light to an even longer through Schoeffel double monochromators. Temporal par- waveleng.th yellow-red. but even smaller amounts of the ameters were controlled with electronic shutters (Vincent -tiOnm light would have been required to cancel the yel- Uniblitz) which provided nearly square-wave pulses either lowness. making the task difficult for the observer at low 17 msec or I set in duration. Intensity was controlled with luminance levels. Therefore 625 nm was chosen as one pri- calibrated Wrattcn neutral density wedges. Radiance mary since its mixture with 170 nm resulted in a unique red measures were made with a photodiode (United Detector with good chromatic saturation and measurements could Technology Pin-lo) and illuminance at %Onm was esti- be obtained at relatively low luminance levels with little mated with a MacBeth Illuminometer and the method of difficulty. The luminance of the 615 nm light was fixed at a Westheimer (1966). A field stop provided a stimulus approx giren level and the luminance of the 170 nm light was 0.6 in diameter. An adjustable bite bar was used to hold varied with the staircase procedure described above. The the observer rigidly in position and fixation was controlled starting points of the two staircases were 1.8 log units apart with four small. dim fixation points arranged in a diamond and successive step sizes in the staircase were 0.4. 0.2. and shape. The stimulus was presented at the intersection of 0.1 log units. imaginary lines connecting the horizontal and vertical One female. and two males. including author AN. served pairs. The observer was instructed to fixate the imaginary as observers. All had normal and extensive intersection point in order to ensure that the stimulus was experience with the judgment before data presented here presented to the rod free fovea. Two considerations were were collected. involved in the choice of the field size. A small field was desirable to ensure that the stimulus was presented to a RESULTS rod free area. in order to avoid possible effects of rod sig- nals on hue at low illuminance levels. Second. it seemed Determination of unique hues with 1 set Hashes are desirable to use a field size small enough to allow the shown in Fig. I and those with 17 msec Rashes are simultaneous presentation of two fields lo the rod free area. shown in Fig. 2. Each plotted point is the mean of at so that data obtained with a hue matching method could be compared to results obtained with the method described Icast three daily means which were obtained by aver- here. (Hue matching data will be reported in a later paper.) aging the last three reversal points in each member of The 36’ field was chosen since it was small enough to meet a pair of staircases for a given condition. The horizon- these considerations. yet large enough to produce trichro- tal lines indicate SO”, confidence intervals based on matic color perception with the procedures used here. between-day variability. Note that the abscissa for Small field tritanopia is generally found with somewhat unique blue, green. and yellow is wavelength. but the smaller fields and appears to occur only with strict conti- abscissa for unique red is the ratio of the illuminances nuous fixation of a long duration stimulus. Bedford and of the 625 and 470nm lights required to produce Wyszecki (19%) have shown that wavelength discrimi- unique red. The unique red data were plotted in this nation in the blue region of the spectrum is nearly as good way to make them more easily comparable to the Hith a 12’ field as it is with a I field if the observer is allowed to scan the Lisual field. and lngling er ~1. (1970) other data. A larger ratio indicates that less blue light ha\e shown that observers are trichromatic even with 3’ is needed to cancel yellowness in the 625 nm light and fields presented to the central fovea if the stimulus is a brief a smaller ratio indicates that more blue is needed. flash. This parallels changes on the \\avelength abscissa Experiments were run in a darkened room with the where an increase in the bvavelength for unique green observer dark-adapted prior to the experiment. On each implies that less blue is needed to cancel the yellow- trial the observer was forced to make a binary decision ness and a decrease in wavelength implies that more about the hue of the test stimulus. For example. if the blue is needed to cancel the yellow. objective was to determine unique yellow or blue. the Consider the results obtained with a I set Hash in observer was asked to respond red or green on each trial to Unique blue indicate whether the stimulus contained some redness or Fig. 1. and unique kellow appear to be greeness. Similarly. for unique red or green. the observer invariant with intensity for all three observers. i&as required to indicate whether the stimulus contained Unique red is clearly not invariant. The 65470 illu- blueness or yellowness. Wavelength was varied according minance ratio must be increased with increasing in- Unique hues are not invariant with brief stimulus durations

480 500 520 I10 590 4.0 - AN

L . I $’ ” ” 470 490 510 590 610 1.9 2.1 2.3 2.5 2.7 WAVELENGTH LoG ‘625 /I 470

Fig. I. Unique hue loci with a I-xc stimulus flash. tensity to maintain unique red for each observer. That indicating that more blue is needed to cancel out yel- is, a smaller proportion of blue light is needed to lowness. At the highest intensity there is a suggestion cancel the yellowness in the 625 nm light as intensity of a reversal of this trend. To maintain unique red as is increased. Unique green also appears to vary with intensity is increased, the proportion of blue light intensity for each observer though the trend is smaller must be decreased in order to just cancel yellowness and not consistent across observers. For DB and EA in the 625 nm light. wavelength must be increased to obtain unique green The dashed lines in Fig. 2 indicate the approximate at the lowest intensity. For AN wavelength must be spectral locations of the unique hues with I set flashes increased as intensity is increased from Fig. I. It appears that use of the shorter Next. consider the results obtained with the duration has not dramatically shifted the locations of 17 msec flash in Fig. 2. All four unique hues shift or the unique hues, implying that the relative weighting lack invariance with intensity. The only case where of the. input signals from cone mechanisms to the there is not a significant shift is unique blue for opponent mechanisms has not been much affected by observer DB. but the trend in DB’s unique blue data stimulus duration. is similar to that of the other observers. The shifts in unique blue and yellow appear to be small in com- DISCUSSION parison with the shifts of unique red and green. Also. note that the shifts for unique red and green are in The results described above show that the invar- opposite directions. As intensity is increased, wave- iance or lack of invariance of unique hues is depen- length must be decreased to maintain unique green, dent on stimulus duration. With I-set flashes. unique 1130 ALLEN L. NAGY

II I:3 1.5 17 I9 4.0- EA Y

480 500 520 4Ot AN i

30.

2’o :T

‘”io., j,i I{, .,1 6io ,\ ,;g’21,2;52;5i7i 490 510 WAVELENGTH LoG ‘625&l

Fig. 2. Solid lines indicate unique hue loci with a 17 msec stimulus flash. Dashed lines indicate unique hue loci with a I-set flash from Fig. I. blue and unique yellow are quite invariant with gest that the transition occurs at durations shorter changes in intensity. Unique green is approximately than 300 msec. invariant but unique red is clearly not invariant. It might be argued that temporal summation These results agree with previous experiments by Lar- should be taken into account in comparing results imer et al. (1974. 1975) with I-set flashes and are con- obtained with the 17 msec flash to those obtained sistent with their conclusion that cone signals input to with the 1-see flash. It is clear that the visual system the red/green ‘opponent mechanism are linear with integrates energy over some period of time. and thus intensity while those input to the yellow/blue mechan- the relevant comparison should be made on a scale of ism are not. Purdy’s early experiments (1931) with energy per unit of integration time, if the integration uncontrolled stimulus duration also produced similar time is longer than 17 msec. For example, Barlow’s results suggesting that this conclusion may hold for experiments on increment thresholds for rods suggest longer stimulus durations as well. However, when a that the visual system integrates energy over approxi- 17 msec stimulus duration is used. all four unique mately 1GOmsec at thresholds levels (see Barlow, hues show consistent departures from invariance. im- 1958). If this integration time applied to suprathresh- plying that cone signals input to both opponent old stimuli as well, the the I-set stimulus would be mechanisms are non-linear functions of intensity. The more effective than the 17 msec stimulus by a factor of transition from short flash behavior to long-flash be- 100/17 or 0.76 log units. The data obtained with the havior has not been determined in these experiments, 17 msec flashes should then be slid downward on the but results reported by Nagy and Zacks (1977) sug- log Troland scale by 0.76 log units for comparison Unique hues are not invariant with brief stimulus durations 1431 with the data obtained with the I-set flash. However, Two possible sources of non-linearity in the short there is clear evidence that the integration time for flash case are the neural links between the receptor suprathreshold stimuli is much shorter than the integ- and the site of opponent cancellation, and the interac- ration time at threshold energy levels. Aiba and tion of the cone signals at the site of cancellation. The Stevens (1964) have found that the interval of com- signals generated in cones presumably have to pass plete temporal summation of energy for the bright- through some anatomical elements before they reach ness of suprathreshold flashes is approximately the site of opponent cancellations. Thus the non- IOmsec for a dark-adapted observer at photopic linearity may occur in a pathway between the recep- levels. If integration time is taken to be only 10msec. tor and this site. data obtained with 17 msec, and I-set flashes are di- The second possibility suggests that the signals rectly comparable without shifting on the log Troland reaching the site of cancellation are linear with inten- scale since both stimuli are longer than the integ- sity but they interact non-linearly. possibly as a result ration interval. Still, it might be argued that the of the briefness of the signal and differences in arrival results obtained by Aiba and Stevens apply only to time at the site of cancellation. There is evidence that the brightness or luminance mechanism, and that the latency of response of a cone mechanism is related temporal summation in the color coding mechanism to its strength of excitation (Alpern, 1954; Drum, should clearly occur over a longer interval than in the 1977) and also that the latency may be different for brightness mechanism (see for example, Regan and different mechanisms (Green, 1969; delange, 1958; Tyler, 1971). The relevant integration interval for the Kelly, 1974; Walraven and Leebeek, 1964; Cicerone, results presented above would appear to be the inter- 1974). Thus, it is possible that brief signals from the val of complete summation for suprathreshold signals individual cone mechanisms arrive at the opponent input to the opponent mechanisms. At presenf no cancellation site at slightly different times and are estimate of this interval appears to exist in the litera- summed or differenced non-linearly as a result. How- ture. so no attempt has been made to shift the curves ever, at present neither of these explanations seems on the basis of an integration argument. In view of capable of reconciling long flash additivity and invar- the fact that the integration time for suprathreshold iance of unique hues with non-linearities found in signals appears to be fairly short, at least for supra- physiological recordings from vertebrate receptors threshold brightness signals. it seems unlikely that with long flashes. taking integration time into account could explain the difference between results obtained with short and Inputs to red/green and yellowjhlue mechanisms long flashes by shifting them to non-overlapping Some inferences can be drawn about the nature of ranges. Even if the short flash data are shifted down- the input signals to the opponent mechanisms from ward by 0.76 log units, which corresponds to an inte- the short flash unique hue data. Over the lower range gration time of lOOmsec, there is still a substantial of intensitites the shifts of unique yellow and unique range of overlap between the short flask curves and blue both indicate that the input to greeness from the the 1 set curves and the results are clearly different at middle wavelength cones grows at a slightly greater comparable levels. Therefore, other possible explana- rate than the input to redness from the short and long tions of the short flash non-linearity are discussed wavelength cones as intensity in increased. At the below. highest intensities of unique yellow the relationship appears to reverse and the input ‘to redness grows Possible sources of non-linearity faster than the input to greeness. The shifts of unique It is perhaps not surprising that the function relat- green and unique red are in opposite directions. That ing response to intensity is dependent upon stimulus is, the shift of unique green suggests that the input to duration, for such a dependence has been shown in yellowness grows at a greater rate than the input to both physiological recordings from vertebrate recep- blueness while the shift of unique red suggests the tors (Fain and Dowling, 1973; Baylor et al. 1974; opposite relationship. This result implies that the Boynton and Whitten, 1970; Normann and Werblin, input to yellowness from the middle wavelength cones 1974) and in human psychophysical experiments on grows at a greater rate than the input to blueness brightness (Mansfield, 1973). What is surprising is from the short wavelength cones, which in turn grows that the nature of this dependence seems to be very at a greater rate than the input to yellowness from the different for color signals on the one hand, and recap- long wavelength cones. tor and brightness signals on the other. Unique hue experiments suggest that signals input to the oppo- SUMMARY nent mechanisms are not linear with short flashes but more nearly linear with long flashes, at least for the The results imply that the properties of signals red/green mechanism. The experiments on receptors input to the opponent mechanisms are to some degree and brightness cited above suggest that signals are dependent on stimulus conditions. This finding im- nearly linear with short flashes and non-linear with poses some limitation on the generality of opponent long flashes. mechanism properties measured under the more typi- If conclusions drawn from the vertebrate receptor cal experimental conditions of color vision experi- recordings apply to human receptors, we are left with ments. Hopefully these limitations are not severe and the problem of explaining the sour& of the non- occur only under extieme conditions such as the very linearity implied by the short flash unique hue data brief flashes used here. Un!il these limits are further and also how the receptor non-linearity is cancelled explored some care should be taken in generalizing out for the red/green mechanism in the long flash properties determined under one set of conditions to case. another set of conditions. I-l52 ALLEZ L. NAG\

The dependency of the intensity-response function Green D. G. (1969) Sinusoidal Ricker characteristics of the on stimulus duration implied by the unique hue data color-sensitive mechanisms of the eye. Vision Rex 9. is consistent with evidence from phys~o~o~cai record- 591-601. ings from receptors and psychophysical experiments Ingling C. R.. Scheibner H. M. 0. and Boynton R. Xl. (1970) Color naming of small fovea1 tields. Vision Rex on brightness. However, the nature of the relationship 10. 501-51 I. between the response function and stimulus duration Jameson D. and Hurvich L. (1955) Some quantitative impiied by the unique hue data appears to be nearly aspects of an opponent colors theory--II. Brightness. opposite that found in other types of experiments. saturation. and hue in normal and dichromatic vision. J. This raises some perplexing problems that remain to opt. Sac. Am. 45, 602616. be resolved. Kelly D. H. (1974) Spatio-temporal frequency character- istics of color vision mechanisms. J. opt. Sot. Am 64. 98 3-990. ctcknonlrdyrmmrs-I would like to thank Dr David Krantz D. H. (1975) Color measurement and color theory Krantz, whose advice and support were invaluable in all -II. Opponent colors theory. J. marh. Ps.vcho/. 12. phases of this work: and Dr R. M. Boynton for his com- 304-3’7. ments on a earlier version of this manuscript. Lange H. de (1958) Research into the dynamic nature of the human fovea-shift in brightness and delay in color perception. J. opt. Sot. Am. 4, 784-789. _ Larimer J.. Krantz D. H. and Cicerone C. M. 11974) REFERENCES Opponent-process additivity: redigreen equilibria. Y&ion Res. 14. 1127-1140. Aiba T. S. and Stevens S. S. (1964) Relations of brightness Larimer J.. Krantz D. H. and Cicerone C. M. (1975) to duration and luminance under light- and dark-adap Opponent-process additivity-II. Yellow/blue equilibria tation. Vision Res. 4, 391-401. and non-linear models. vision Res. 15, 723-732. Alpern M. (1954) The relation of visual fatency to intensity. ,ManstieId R. 1. W. (1973) Brightness function: effect of area A.ltf.A. Arch. Oprh. 51, 369-374. and duration. 1. Opt. Sot. Am. 63, 913-920. Barlow H. B. (1958) Temporal and spatial summation in Maxwell J. C. (1856) Experiments on color as perceived by human vision at different background intensities. J. Ph.v- the eye with remarks on colour blindness. Trans. R. Sot. siof.. Land. 141, 337-350. Edinb.. reprinted in The Scientific Papers of JamesClerk Baylor D. A.. Hodgkin A. L. and Lamb T. (1974) The Maxweell (Edited by Niven W. D.). Dover, New York. electrical response of turtle cones to flashes and steps of Nagy A. L. and Zacks J. L. (1977) The effects of psycho- light. J. Physiol.. Land. 242, 759-791. physical procedure and stimulus duration -in- the Bedford R. E. and Wyzsecki G. W. (1958) Wavelength dis- measurement of Bezold-Brucke hue shifts. Vision Rex crimination for point sources. f. opt. Sot. Am. 48. 17. 193-200. 129-135. Normann R. A. and Werbiin F. S. (1974) Control of retinal Boynton R. M and Whitten D. N. (1970) Visuai adaptation sensitivity--I. Light and dark adaptation of vertebrate in monkey cones: Recordings of late receptor potentials. rods and cones. J. gen. Ph_vsio/. 63, 37-61. Science, 5.X 1423-I426. _ Pierce C. S. (1877) Note on the sensation of color. Am. J. Cicerone C. M. 0. (1974) Theory of luminance detection Sci. Ser. 3. 13. 247. and measurement’ of the impulse responses and fre- Purdy D. M. (1931) Spectral hue as a function of intensity. quency responses of the color mechanisms of the eye. Am. J. Psycho/. 43. 541-559. Mich. marh. Psychoi. reck. Rep. 74-6. Regan D. and Tyler C. W. (1971) Temporal summation Cornsweet 1. N. (1962) The staircase method in psycho- and its limit for wavelength changes: an analog of physics. Am. J. Psychof. 75. 485-491. Bloch’s Law for color vision, J. opr. Sot. Am. 61, 1414- Drum B. A. (1977) Cone interactions at high flicker fre- 1421. quencies: Evidence for cone latency differences? J. opt. Savoie R. S. (1973) Bezold-Briicke effect and visual non- Sot. Am. 67, l&X-1603. linearity. 1. opt. Sot. Am. 63, t253-1261. Fain G. L. and Dowiing J. E. (1973) Intracellular recotd- Walraven P. L. (1961) On the Bezold-Briicke phenomenon. ings from single rods and cones in the mudpuppy retina. J. opt. Sac. Am. 51, 1113-1116. Science, ?;.Y. 180. 1178-l 181. Walraven P. L.-and Leebeek H. J. (1964) Phase shift of Genter C. R.. Geri G. A. and Kandel G. L. (1978) sinusoidally alternating colored stimuli. J. opt. Sac. Am. Luminance dependent shifts in unique yellow. Abstract 54, 78-82. in Supplement to Invest. Opihal. and Visual Science, April, Westheimer G. (1966) The Maxwellian view. Vision Res. 6, 1978. 669-682.