Contrast discrimination with pulse trains in pink noise

G. Bruce Henning, Chris M. Bird, Felix A. WichmannJournal of the Optical Society of America A 2002 · DOI

TL;DR

This paper investigates how our visual system combines information from different spatial frequencies, particularly when dealing with complex stimuli like pulse trains. They found that while adding noise can improve the detection of pulse trains by making their components equally detectable, this benefit disappears when discriminating contrasts in the presence of a pedestal, suggesting that internal noise and signal correlations play a crucial role in limiting performance.

What Problem Does This Paper Address?

Understanding how our visual system processes spatial information, especially at low contrasts near detection thresholds, is a fundamental challenge in vision science. For decades, researchers have used sinusoidal gratings to probe the mechanisms of contrast detection and discrimination, often assuming the visual system operates with independent spatial frequency channels. However, real-world stimuli are rarely simple sinusoids. This paper tackles the question of how the visual system handles more complex stimuli, specifically pulse trains, which contain multiple spatial frequency components. They aim to understand how information from these components is integrated and what limits our ability to discriminate fine differences in contrast, particularly in the context of "Weber's law" (the idea that the just-noticeable difference in stimulus intensity is proportional to the stimulus intensity). A key question is whether performance with complex stimuli is simply the sum of performance with its individual components, or if there are more sophisticated integration mechanisms at play.

Methods

The researchers employed a two-alternative forced-choice (2AFC) psychophysical paradigm to measure contrast detection and discrimination. They used custom-generated stimuli:

  • Sinusoidal Gratings: Standard stimuli used as a baseline for comparison.
  • Pulse-Train Gratings: These are periodic patterns of narrow lines. Theoretically, an ideal pulse train contains harmonics of all orders at equal contrast. The authors created approximations of these using a display with limitations on pixel size and luminance.
  • Pink Noise: Broadband noise with a power spectral density inversely proportional to spatial frequency. This noise was carefully designed to equalize the detectability of the individual frequency components within the pulse train.

The experiments involved:

  1. Detection: Participants identified the presence of a signal grating (either sinusoidal or pulse-train) against a uniform background.
  2. Discrimination: Participants discriminated between two intervals, one containing a "pedestal" grating (a grating of 15% contrast) and the other containing the pedestal plus a signal grating. This is essentially a contrast increment detection task.

They systematically varied the presence and type of stimuli (sinusoids vs. pulse trains) and the presence of masking pink noise. Psychometric functions were fitted with Weibull functions to determine thresholds and slopes.

Datasets

Not discussed in the paper. The stimuli were generated by the researchers using custom software (MATLAB) and displayed on a calibrated monitor.

Key Insights and Ideas

  • Pulse Trains vs. Sinusoids in Detection: When presented without noise, a pulse train was not significantly more detectable than its most detectable single-frequency component, even though it contained multiple harmonically related components. This suggests that simple probability summation (where multiple independent cues lead to improved detection) wasn't fully exploited.
  • Noise Equates Detectability: The addition of pink noise, designed to equalize the detectability of the pulse train's components, dramatically improved pulse train detection. This suggests that the unequal detectability of components was a major limitation, and that observers can benefit from multiple cues when they are made equally salient.
  • Noise Fails to Help Discrimination: Crucially, when a 15% contrast pedestal was present, the addition of pink noise did not improve contrast discrimination performance for the pulse train relative to its sinusoidal components. This is a key finding, implying that the mechanisms limiting discrimination are different from those limiting detection.
  • Internal Noise and Correlation: The lack of improvement in discrimination with noise suggests that either the "internal noise" (noise inherent to the visual system) is highly correlated across spatial frequency channels, or that the signal itself becomes highly correlated with the pedestal at higher contrasts, preventing independent processing of components. This challenges simple models of independent channels.

Key Results

  • Detection (0% Pedestal):
    • Pulse train detection was only slightly better than its most detectable component without noise.
  • With pink noise, pulse train detection improved by approximately a factor of 5 compared to its components.
  • Discrimination (15% Pedestal):
    • With a 15% pedestal, the pulse train's contrast discrimination performance was very similar to that of its individual sinusoidal components.
  • Adding pink noise did not improve pulse train discrimination performance relative to its components in the presence of the pedestal.
  • "Super Train" Experiment: A synthesized stimulus ("super train") where components were adjusted to be equally detectable showed a factor of 2 improvement in detection over single sinusoids, supporting the idea of probability summation. However, the psychometric function slopes were not steeper, suggesting limitations in how this summation is utilized.

Limitations and Future Work

  • Display Limitations: The authors acknowledge that they could not create an "ideal" pulse train due to display limitations (pixel size, dynamic range), which affected the maximum achievable contrast of components.
  • Internal Noise Source: While they infer the presence of internal noise and its potential correlation, the exact source of this noise (e.g., neural variability, gain control mechanisms) remains an open question. They speculate about eye movements or early gain control mechanisms.
  • Future Directions: The paper suggests that further investigation into the nature of internal noise and its correlation across channels is needed. Understanding how this internal noise interacts with external noise and signal properties is crucial for developing more complete models of early visual processing. The findings also point towards the need for models that can account for both detection and discrimination performance, especially concerning the role of pedestals and internal noise.

Summary updated

Abstract

Detection performance was measured with sinusoidal and pulse-train gratings. Although the 2.09-cycles-per-degree pulse-train, or line, grating contained at least eight harmonics all at equal contrast, it was no more detectable than its most detectable component. The addition of broadband pink noise designed to equalize the detectability of the components of the pulse train made the pulse train approximately a factor of 4 more detectable than any of its components. However, in contrast-discrimination experiments, with a pedestal or masking grating of the same form and phase as the signal and with 15% contrast, the noise did not affect the discrimination performance of the pulse train relative to that obtained with its sinusoidal components. We discuss the implications of these observations for models of early vision, in particular the implications for possible sources of internal noise.
BibTeX
@article{henning2002contrast,
  author = {G. Bruce Henning and Chris M. Bird and Felix A. Wichmann},
  title = {{Contrast discrimination with pulse trains in pink noise}},
  journal = {Journal of the Optical Society of America A},
  year = {2002},
  doi = {10.1364/josaa.19.001259},
}
Figure