CSFs represent the human visual system’s ability to detect contrast variations and have found important applications in engineering, where they can be used to optimise designs to cater to human perceptual limits. A comprehensive CSF model requires consideration of stimulus parameters, including spatial and temporal frequencies, luminance, and colour, among others. Despite an extensive collection of contrast sensitivity measurements in the literature, no current model covers the full stimulus parameter space.
The inception of ModelFest project more than two decades ago marked a pivotal moment in the research towards a unified visual detection modelling approach (Carney et al., 1999). The ambitious initiative laid the foundation for integrating diverse stimuli measurements under a cohesive framework. For CSF modelling specifically, the physiological models from Barten (1999) and the analytical Pyramid of Visibility models (Watson and Ahumada, 2016; Watson, 2018) have been key advancements in the research area. Our work on CSF during the last few years has been inspired by these approaches.
The study by Wuerger et al. (2020) marks the beginning of our work on modelling CSF. This work emphasises the importance of colour modulations alongside spatial frequency and luminance. The work by Mantiuk et al. (2020) extends this framework to include background chromaticity effects and compared cone contrast and post-receptoral contrast encodings. In the proposed stelaCSF model (Mantiuk et al., 2022), achromatic contrast sensitivity was modelled by synthesising 11 distinct datasets. This work aimed for a robust and generalised model that could predict sensitivity across spatial and temporal frequencies, luminance, size and eccentricity.
The latest iteration of our work, the castleCSF model (Ashraf et al., 2024), combines the strengths of preceding studies and uses datasets from 18 studies to predict sensitivity to spatial and temporal frequencies, any arbitrary contrast modulation direction in the colour space, mean luminance and chromaticity of the background, eccentricity, and stimulus area with a mean error of 3.59 dB. One major feature of our model, distinguishing it from other current works, is its use of the same set of parameters to explain data from 18 different studies, demonstrating its robustness and generalisability. This model offers insights into the mechanisms affecting contrast sensitivity for different stimulus parameters using an analytical modelling approach informed by known behaviour of physiological components governing contrast sensitivity.
