Chromatic detection and discrimination analyzed by a Bayesian classifier.

Detection and threshold-level discrimination of Gabor patches were studied under the conditions of noise masking, in an attempt to isolate 'higher-order' or nonclassical color mechanisms. Detection contours in the equiluminant plane of cone contrast space were measured by varying test chromaticity in the presence of chromatic masking noise. Three equiluminant noise directions were used, in separate experiments. In the discrimination experiment, observers had to discriminate between pairs of stimuli that were fixed at their masked threshold contrasts. A Bayesian color classifier model was used to analyze the discrimination data, with no free parameters. There was no evidence of nonclassical color mechanisms in either the detection or discrimination data.

Regional cerebral correlates of global motion perception: evidence from unilateral cerebral brain damage.

We used a psychophysical task to measure sensitivity to motion direction in 50 stroke patients with unilateral brain lesions and 85 control subjects. Subjects were asked to discriminate the overall direction of motion in dynamic stochastic random dot displays in which only a variable proportion of the spots moved in a single direction while the remainder moved randomly. Behavioural and neurophysiological evidence shows that the middle temporal (MT/V5) and middle superior temporal (MST) areas in the macaque monkey are indispensably involved in the perception of this type of motion. In human subjects too, lesions in the same region disrupt performance on this task. Here we assessed more extensively the correlation between direction sensitivity for global motion and the anatomical locus of the lesion. Thresholds for perceiving the direction of global motion were impaired in the visual field contralateral to the lesion in patients with lesions in the occipitoparietal and parietotemporal areas involving the human analogue of areas MT/V5 and MST, but not by lesions in the occipito-temporal or anterior frontal areas. Patients with lesions involving the anterior temporal or parietal lobes displayed poor performance for stimuli presented in either visual field, which is consistent with the large and bilateral receptive fields in these areas in monkeys. The perception of global motion was also more impaired in the centripetal than the centrifugal direction in the hemifield contralateral to the MT/V5 lesion. Surprisingly, thresholds were normal in all patients when the displays contained static but not dynamic visual noise, suggesting that their deficit reflects an inability to filter out dynamic noise. Although frequent repeated testing of some patients whose lesion involved the human homologue of MT was accompanied by an improvement in performance, this was no greater than in other patients who received training on different motion tasks.

ON and OFF S-cone pathways have different long-wave cone inputs.

Three experiments compared thresholds for S-cone increments and decrements under steady and transient adaptation conditions, to investigate whether stimuli of both polarities are detected by the same cone-opponent psychophysical mechanism. The results could not be accounted for by a standard model of the S-cone detection pathway [Polden & Mollon (1980) Proceedings of the Royal Society of London, B, 210, 235-272]. In particular, a transient tritanopia detection paradigm that measured threshold elevation following the offset of long-wavelength fields produced different field sensitivities for S-cone increment and decrement tests. The decrement field sensitivity function was shifted to shorter wavelengths relative to the increment function. L-cone opponency is apparently stronger for S-cone increments than for decrements. The most plausible substrates of the two different psychophysical detection mechanisms are the ON and OFF channels. The results suggest that S-ON (bistratified) and S-OFF ganglion cells receive different relative amounts of L- and M-cone input.

Chromatic masking in the (delta L/L, delta M/M) plane of cone-contrast space reveals only two detection mechanisms.

The post-receptoral mechanisms that mediate detection of stimuli in the (delta L/L, delta M/M) plane of color space were characterized using noise masking. Chromatic masking noises of different chromaticities and spatial configurations were used, and threshold contours for the detection of Gaussian and Gabor tests were measured. The results do not show masking that is narrowly-selective for the chromaticity of the noise. On the contrary, our findings suggest that detection of these tests is mediated only by an opponent chromatic mechanism (a red-green mechanism) and a non-opponent luminance mechanism. These results are not consistent with the hypothesis of multiple chromatic mechanisms mediating detection in this color plane [1].

Rapid sensitivity changes on flickering backgrounds: tests of models of light adaptation.

To investigate mechanisms underlying sensitivity changes that are capable of following rapid variations in intensity of the background field, we measured the threshold radiance needed to detect a 2-ms probe flash presented at various phases relative to a sinusoidally flickering background. The temporal frequency, mean luminance, and modulation of the background were systematically varied. The sensitivity change consisted of two components: a phase-insensitive increase in threshold that occurs at all the phases of the background field (a change in the dc level of the threshold), and a phase-dependent variation in threshold. Both components can reliably be measured at temporal frequencies up to approximately 50 Hz. On a 30-Hz background, the threshold varied with phase over roughly 0.5 log unit within a half-cycle (17 ms). For background flicker rates of 20-40 Hz the probe threshold increased with increasing instantaneous background radiance, following a typical threshold-versus-radiance template, and approaching Weber-law behavior during the peak of the background flicker. This pattern of threshold elevation was measured at mean background illuminances from 580 to 9100 Td (trolands), with the dimmer backgrounds being slightly less effective in producing threshold elevations. The measured increase in the dc level commenced as soon as the modulation of the background flicker began, and the amount of threshold elevation followed the envelope of the background flicker, ruling out modulation gain control explanations for the change in sensitivity on flickering backgrounds. The threshold elevations measured on a 30-Hz, 25% modulation background were lower than those measured on a 30-Hz, 100% modulation background at all phases. The measured changes in threshold with changes in background modulation rule out all adaptation models consisting of a multiplicative and a subtractive adaptation processes followed by a single, late, static nonlinearity.

Contributions of human long-wave and middle-wave cones to motion detection.

1. It has been suggested that motion may be best detected by the luminance mechanism. If this is the most sensitive mechanism, motion thresholds may be used to isolate the luminance mechanism and study its properties. 2. A moving (1 cycle deg-1), vertical, heterochromatic (red-plus-green), foveal grating was presented on a bright yellow (577 nm wavelength) field. Detection and motion (direction identification: left versus right) thresholds were measured for different amplitude ratios of the red and green components spatially summed in phase or in antiphase. Threshold contours plotted in cone-contrast co-ordinates (L',M') for the long-wave (L) and middle-wave (M) cones, revealed two motion mechanisms: a luminance mechanism that responds to a weighted sum of L and M contrasts, and a spectrally opponent mechanism that responds to a weighted difference. 3. Detection and motion thresholds, measured at 1-4 Hz, were identical for luminance gratings, having equal cone contrasts, L' and M', of the same sign. For chromatic gratings, with L' and M' of opposite sign, motion thresholds were higher than detection thresholds. A red-green hue mechanism may mediate chromatic detection, and a separate spectrally opponent motion mechanism may mediate motion. 4. The red-green hue mechanism was assessed from 1 to 15 Hz with an explicit hue criterion. The detection contour had a constant slope of one, implying equal L' and M' contributions of opposite sign. For motion identification, L' and M' contributed equally at 1 Hz, but the M' contribution was attenuated at higher velocities. 5. The cone-contrast metric provides a physiologically relevant comparison of sensitivities of the two motion mechanisms. At 1 Hz, the spectrally opponent motion mechanism is approximately 4 times more sensitive than the luminance mechanism. As temporal frequency is increased, the relative sensitivities change so that the luminance mechanism is more sensitive above 9 Hz. 6. The less sensitive motion mechanism was isolated with a quadrature phase protocol, using a pair of heterochromatic red-plus-green gratings, counterphase flickering in spatial and temporal quadrature phase with respect to each other. One grating was set slightly suprathreshold and oriented in cone contrast (L',M') so as to potentiate a single motion mechanism, the sensitivity of which was probed with the second grating, which was varied in (L',M'). This allowed us to measure the motion detection contour of the less sensitive luminance mechanism at low velocities.(ABSTRACT TRUNCATED AT 400 WORDS)

Temporal properties of the red-green chromatic mechanism.

The temporal properties of the red-green chromatic mechanism were studied with red and green equiluminant flashes of 1 deg diameter presented in the center of a bright (800-3000 td) yellow adapting field. A subthreshold 200 msec red or green flash makes an immediately subsequent, suprathreshold yellow luminance flash appear tinged with the complementary color. The chromatic flash also makes a subsequent chromatic flash of the same hue harder to detect and identify, and makes a flash of the opposite hue easier to detect and identify. These results indicate that the response of the red-green mechanism changes polarity during its time-course, suggesting that the chromatic temporal impulse-response function has a negative lobe. Pairs of chromatic pulses were used to estimate the shape of the chromatic impulse-response. The estimated impulse-response function has a zero crossing near 90 msec, followed by a long, shallow negative lobe. We also measured threshold-duration functions; the critical duration for the chromatic and luminance flashes is about 95 and 45 msec, respectively. Chromatic sensitivity (measured in cone contrast units) is 10 times greater than luminance sensitivity for long durations, and is 3 times greater for all durations less than 45 msec.

The time-course of chromatic facilitation by luminance contours.

The threshold for detecting an equiluminant chromatic spot is approximately halved by the presentation of a coincident, suprathreshold luminance pedestal flash. The dynamics of this facilitation were studied by varying the duration and temporal asynchrony of the chromatic test flash and luminance pedestal. Facilitation occurs in a narrow temporal window near the chromatic test presentation; masking may occur when the pedestal is temporally displaced from the test by longer times. The mechanism producing facilitation lags behind the chromatic signal by at least 20 msec.

The foveal color-match-area effect.

Rayleigh matches for foveal, temporally alternating fields showed only a small increase in the log green/red matching ratio (0.03--average of 10 observers) as the field was decreased from 116 to 19 min arc. This is consistent with only a small, 10%, change in photopigment density or lengthening of the cone outer segments in the central fovea. The change in matches with field size is considerably less than reported in several previous studies.

Separable red-green and luminance detectors for small flashes.

Detection contours were measured in L and M cone contrast coordinates for foveal flashes of 200 msec duration and 2.3, 5, 10 and 15 min arc diameter on a bright yellow field. The test flash consisted of simultaneous incremental and decremental red and green lights in various amplitude ratios. At all sizes, the most sensitive detection mechanism was not a luminance mechanism, but rather a red-green mechanism that responds to the linear difference of equally weighted L and M cone contrasts, and signals red or green sensations at the detection threshold. Both temporal and spatial integration were greater for red-green detection than luminance detection. A coincident, subthreshold, yellow flash (a luminance pedestal) did not affect the threshold of the red-green mechanism. Such a pedestal is a sum of equal L and M cone contrast--it represents a vector parallel to the red-green detection contour and thus is expected not to stimulate directly the red-green mechanism. When suprathreshold, the coincident pedestal facilitated chromatic detection by approximately 2x at all tested sizes; intense pedestals did not mask chromatic detection. This insensitivity to intense luminance pedestals further indicates that the red-green mechanism has fixed spectral tuning with balanced opponent L and M contrast inputs. This view of fixed spectral weights contrasts with the "variable tuning hypothesis", which postulates that the weights change with spatial-temporal variations in the test stimulus.(ABSTRACT TRUNCATED AT 250 WORDS)

Colour is what the eye sees best.

It has been argued by Watson, Barlow and Robson that the visual stimulus that humans detect best specifies the spatial-temporal structure of the receptive field of the most sensitive visual neurons. To investigate 'what the eye sees best' they used stimuli that varied in luminance alone. Because the most abundant primate retinal ganglion cells, the P cells, are colour-opponent, we might expect that a coloured pattern would also be detected well. We generalized Watson et al.'s study to include variations in colour as well as luminance. We report here that our best detected coloured stimulus was seen 5-9-fold better than our best luminance spot and 3-8-fold better than Watson's best luminance stimulus. The high sensitivity to colour is consistent with the prevalence and high colour contrast-gain of retinal P cells, and may compensate for the low chromatic contrasts typically found in natural scenes.

Peripheral chromatic sensitivity for flashes: a post-receptoral red-green asymmetry.

Thresholds of luminance and red-green chromatic flashes (200 msec) were measured on a yellow adapting field in the fovea and periphery (up to 12 degrees eccentricity for 1 degree flashes and 21 degrees eccentricity for 2 degrees flashes). Chromatic sensitivity (in cone contrast coordinates) is about 7 times higher than luminance sensitivity in the fovea but falls faster with eccentricity, so that luminance and chromatic sensitivities are similar at eccentricities of 20 degrees or less. At eccentricities greater than about 14 degrees, there is a clear asymmetry wherein green chromatic flashes are considerably less detectable than red ones. By measuring complete detection contours for many ratios of incremental and decremental red and green flashes, we isolated the red and green chromatic detection mechanisms, and demonstrated that the red-green asymmetry is not a property of the L- or M-cone response per se, but rather is a property of the post-receptoral, chromatic mechanisms. The peripheral luminance and chromatic mechanisms could be further separated with a suprathreshold luminance flash (a pedestal), since an intense pedestal masks coincident luminance test flashes but facilitates the chromatic flashes. The luminance pedestal approximately linearizes the chromatic detection function (the psychometric function).

Detection uncertainty and the facilitation of chromatic detection by luminance contours.

A suprathreshold luminance flash (1 degree, 200 msec) on a large uniform yellow field facilitates detection of a coincident (1 degree, 200 msec) red or green equiluminant flash and approximately linearizes the psychometric function for detecting the chromatic flash. The facilitation is produced by the suprathreshold contour created by the luminance flash. We tested whether the contour facilitates detection by reducing spatiotemporal uncertainty in detecting the chromatic flash. Uncertainty increase false alarms, and this effect can be factored out by correcting yes-no psychometric functions for guessing. Uncertainty also alters the shape of the receiver operating characteristic. Measurements of yes-no psychometric functions and receiver operating characteristics do not support the uncertainty reduction hypothesis.

Temporal phase response of the short-wave cone signal for color and luminance.

A chromatic discrimination paradigm was used to measure the temporal phase of the S (short-wave cone) signal relative to the L--M (long-wave cone minus middle-wave cone) signal. Suprathreshold equiluminant red-green flicker that stimulates the L--M mechanism was presented on a steady, intense yellow-green adapting field. Violet flicker that stimulates the S cones was added to the red-green flicker at different temporal phase angles, and the violet modulation depth was varied to achieve a chromatic discrimination threshold. A template was fitted to the data relating thresholds to phase: the location of the template symmetry axis showed that the S signal lagged L--M by about 75-90 degrees at 10 Hz. This is about one half the phase lag obtained for luminance or motion discrimination. The phase discrepancy shows that there are separate luminance and chromatic mechanisms receiving S cone inputs. The hue of the flicker in the present study varied strongly with phase angle, with the positive and negative excursions of the S cone signal producing a reddish-blue and greenish-yellow, respectively, and these colors combined with the reddish and greenish hues produced by the L--M signal. The observed phase shift, and measured color appearance of the combined flicker, account for the colors seen on a radially segmented disk of Munsell hues when rotated: the colors differ strikingly depending on the direction of rotation.

The gap effect revisited: slow changes in chromatic sensitivity as affected by luminance and chromatic borders.

Chromatic discrimination was studied with two half-fields that were either precisely juxtaposed or were separated by a narrow gap. When present, the gap was in some conditions filled with light isoluminant to the test fields. When the fields were juxtaposed, chromatic sensitivity declined with viewing duration. For a discrimination based solely on S cone activity, separating steadily-viewed fields by either a luminance or a purely chromatic gap caused similar enhancements of sensitivity. Neither type of gap had much effect when the fields were flashed. The results may be interpreted as showing that either a luminance or chromatic contour can spatially delimit the two half-fields, thus preventing a slow spatial integration from reducing the discriminability of the two sides of the field.

Analysis of color-matching ellipses in a cone-excitation space.

Color-discrimination ellipses derived from the variability of color-matching data of six observers are analyzed in a normalized constant-luminance cone-excitation space. The analysis shows that the ellipses do not vary significantly in shape with chromaticity, observer, or experimental conditions. The discrimination contours are predictable from the thresholds on the two cardinal axes of this space; these are used to normalize the data at each chromaticity for each observer. Thresholds on these two axes vary with chromaticities, individuals, and experimental conditions in accordance with simple and familiar laws.

Effects of field area and configuration on chromatic and border discriminations.

Visual discrimination was tested with juxtaposed fields that differed only in luminance, S cone excitation, or the ratio of L to M cone excitation. Observers were asked to make discriminations based upon a criterion of color difference or of the perception of a border between fields. The area of the fields was varied either by increasing the length of their common junction or by increasing their widths at constant junction length. For fields larger than 0.2-0.4 deg2, area controls sensitivity. For smaller fields, discrimination may be worse for narrow fields with long common junctions than for wide ones with small common junctions, especially for discriminations that depend upon S cones, in which the field components tend to become spatially integrated. Decreasing viewing time reduces the magnitude of this effect. S cone and L-M channel acuities were estimated at about 7 and 32 c/deg, respectively.

Similarity of normalized discrimination ellipses in the constant-luminance chromaticity plane.

Discrimination steps were measured for three subjects, along oblique axes passing through nine points in a 25 td constant-luminance chromaticity plane. When plotted in a normalized cone-excitation chromaticity diagram, the best-fitting discrimination ellipses for a given subject have approximately the same shape and orientation regardless of the reference chromaticity. Their orientation is consistent with the hypothesis that excitation of B-cones affects the red-green opponent balance, otherwise determined by R- and G-cone excitations, in a manner independent of initial cone-excitation levels. The CIELAB formula predicts an orientation for normalized ellipses in agreement with the data, but it also predicts systematic changes in the ratio of minor to major axes which are not observed experimentally.

Blue cones contribute to border distinctness.

A conclusion reached by Kaiser and Boynton in wavelength discrimination experiments, that differential blue-cone excitation makes a small contribution to the distinctness of borders in small fields, was verified using a rating procedure for borders produced in the La Jolla Analytic Colorimeter.

Graded differential reinforcement: Response-dependent reinforcer amount.

After key pecking had been autoshaped, six pigeons were exposed to a condition in which the duration of grain availability at the end of an 8-second trial depended on the number of responses emitted during the trial (0.25-second access to grain per response). This procedure, called correlated reinforcement, alternated across conditions with the automaintenance baseline in which the 8-second trial terminated with a constant 2.5-second access to grain. Two control procedures were run; in both, the reinforcer durations were yoked to those obtained in the last correlated session. In the yoked control no responses were required, but in the single-response yoked control at least one response was required to receive the yoked duration. The correlated condition maintained response rates above those produced by the two control conditions. These results may be accounted for by differential reinforcement.