Skip to contents

Pairwise mark-discrimination probability matrix for a palette

Source: R/palette_prob_bg.R
palette_prob_bg.Rd

The probability analogue of palette_dist_bg()'s "min-lum" method: for every pair of colors in x, the probability that two size-sized marks (line marks, or point/scatterplot marks) drawn in those colors would be told apart, using the Szafir (2018) mark-discrimination models (see line_discrim_prob() / point_discrim_prob()) instead of a colorblindcheck::palette_dist() distance.

Usage

palette_prob_bg(
  x,
  bgcol = "#FFFFFF",
  cvd = NULL,
  severity = 1,
  thickness = 0.05,
  mark = c("line", "point"),
  metric = 2000,
  lum_k = 100,
  lum_floor = 0.05,
  lum_adjust = FALSE
)

Arguments

x

vector of hex colors (the palette)

bgcol

background color as a hex string. Only used when lum_adjust = TRUE.

cvd

type of color vision deficiency to simulate before computing probabilities, or NULL for none. See colorblindcheck::palette_dist() for supported values.

severity

severity of the CVD simulation, between 0 and 1

thickness

mark size in degrees of visual angle – line thickness when mark = "line", point diameter when mark = "point" – passed to the underlying line_discrim_prob() / point_discrim_prob() model. See their documentation for the (different) validated ranges.

mark

which Szafir (2018) model to use: "line" (line graphs) or "point" (scatterplots). See line_discrim_prob() and point_discrim_prob().

metric

accepted for signature parity with palette_dist_bg(), but unused: this model works directly on CIELAB L*/a*/b*, not on a colorblindcheck::palette_dist() distance.

lum_k, lum_floor

passed to lum_factor() when lum_adjust = TRUE; see that function's documentation for their meaning.

lum_adjust

should the probability be multiplied by lum_factor()? Exploratory and unvalidated – see Details.

Value

n x n matrix of discrimination probabilities as percentages (0-100), with NA on the diagonal

Details

Only "min-lum"'s weakest-link-plus-lightness-penalty idea is mirrored here, since it is the only palette_dist_bg() method with a coherent probability interpretation: "bg-norm" and "bg-norm-cr-spread" are distance-space transforms (squared-distance ratios, log-contrast spread) that don't map onto a [0, 1]-bounded probability, so there is no method argument here.

Unlike palette_dist_bg(), this function does not compute a mark-vs-background discriminability term (the bdiff piece of "min-lum"). Szafir's models were fit on two thin/small marks compared against each other on a field of gray distractors, not a mark against a large uniform background field, so reusing that ΔE2000-based bdiff logic here is not well founded. lum_adjust = TRUE is offered as an exploratory alternative instead (see below) – if a background-visibility floor is wanted for this model, it needs its own justification, not a straight port from the distance version.

Caveats, in addition to those on line_discrim_prob() / point_discrim_prob() (note the two marks have different validated size ranges – 0.05-0.35 degrees for "line", 0.25-2.0 degrees for "point"):

  • lum_adjust = TRUE is exploratory, not validated. It multiplies the Szafir-derived probability by lum_factor(), which was fit on ΔE-based distance data, not probabilities. There is no empirical basis yet for combining the two this way – treat it as a hypothesis to test, not a trustworthy correction. It defaults to FALSE.

See also

line_discrim_prob(), point_discrim_prob()