Contrast sensitivity function

The contrast sensitivity function (CSF) describes how the minimum contrast needed to see a pattern changes with its spatial frequency. Sensitivity is the reciprocal of the threshold contrast. The function characterizes the human visual system across a range of detail sizes, rather than giving a single resolution limit.

Reviewed 21 October 2026. Contrast definitions, measurement methods, perceptual models and the cited headset studies. About review dates.

In a head-mounted display, contrast detection depends on both the observer and the imagery delivered through the headset. Researchers use perceptual measurements and CSF models to connect optical measurements with what users can detect.

Definition and interpretation

The conventional spatial CSF uses sinusoidal gratings: alternating light and dark bands with a smoothly varying luminance. Spatial frequency expresses the number of complete light-dark cycles per degree of visual angle. Lower frequencies describe coarser patterns; higher frequencies describe finer patterns. At each frequency, the experiment determines the contrast at which the pattern becomes just detectable.

For a sinusoidal grating, Michelson contrast is:

C = (L_max - L_min) / (L_max + L_min)

Here L_max and L_min are maximum and minimum luminance. Contrast sensitivity at frequency f is S(f) = 1 / C_threshold(f). A higher sensitivity means that less contrast is needed for detection.

The curve depends on the viewing conditions. In Campbell and Robson's 1968 experiments, sensitivity fell at both high and low spatial frequencies, and increasing the angular size of the visible grating raised low-frequency sensitivity. Their results also showed that different grating waveforms could have different detection thresholds.

Measurement

A test must specify the observer's task and the criterion used to estimate threshold. Conventional procedures can estimate thresholds separately at a set of spatial frequencies. The quick CSF method, introduced by Lesmes and colleagues in 2010, instead updates a probability distribution over candidate curves after each response. It selects both frequency and contrast to maximize expected information gain about the curve's parameters.

The original quick CSF implementation represented the curve with four parameters: peak sensitivity, the frequency of the peak, bandwidth and a low-frequency truncation level. A fitted curve therefore combines the observer's responses with an assumed functional form. The paper also used the area under the log CSF as a summary measure.

Contrast sensitivity testing differs from conventional high-contrast visual acuity testing. In a letter acuity chart, the letters become smaller. The Pelli-Robson chart instead retains a constant letter size while reducing contrast. Its letter-identification score is a contrast-sensitivity measure, rather than a grating CSF sampled across multiple frequencies.

Viewing conditions and models

Contrast sensitivity varies with luminance, stimulus size, position relative to fixation and temporal modulation. A single spatial curve represents a restricted set of these conditions. Mantiuk, Ashraf and Chapiro's 2022 stelaCSF model combined measurements from 11 published studies to model spatial frequency, temporal frequency, eccentricity, luminance and area. It addressed achromatic patterns; chromatic sensitivity was outside its scope.

The authors demonstrated model applications to flicker detection and foveated rendering. Their foveation example simulated how different display sampling densities change the sensitivity-based choice of peripheral sampling levels.

Headset image quality

An optical modulation transfer function (MTF) describes the loss of image modulation at different spatial frequencies. It is distinct from an observer's threshold sensitivity. In a 2024 study, Zhao and colleagues incorporated measured headset MTFs and an external-noise term into a modified Barten contrast-sensitivity model, then compared its predictions with perceptual measurements on a Meta Quest 2.

The study included nine participants and measured right-eye contrast thresholds with the other eye covered. It compared camera arrangements representing pupil rotation, eye rotation and eye rotation with an allowance for angle kappa, the offset between the eye's optical and visual axes. The last arrangement produced the strongest reported correlation with the perceptual measurements. The authors identified the small sample, participants' direct adjustment of contrast and the exclusion of binocular vision as limitations.

Monocular and binocular measurements

Bhansali and colleagues' separate 2024 study used WebXR and A-Frame to measure monocular and binocular contrast detection on a Quest 2. Nine participants viewed Gabor patterns, which combine a sinusoidal pattern with a Gaussian envelope, at nine display locations. The experiments included matched and mismatched interpupillary distance settings.

The authors called their measurement a contrast sensitivity response (CSR) to distinguish it from a CSF of vision without the headset's optical modulation. They found that binocular performance was primarily dominated by the eye with better contrast sensitivity in the tested configurations. This was a test-method study using one headset, rather than a comparison across headset technologies.

Display calibration also matters: input gray-level contrast is not necessarily the same as luminance contrast. The study measured the headset's luminance response to convert between them. It identified possible response bias from manual contrast adjustment and left the influence of virtual viewing distance for future investigation.

Limits of interpretation

A CSF primarily models near-threshold patterns on uniform backgrounds. It does not by itself explain the appearance of clearly visible patterns, contrast masking on patterned backgrounds or visual crowding. Those effects require additional modeling. A sensitivity curve is therefore one component of perceptual image-quality assessment.