Skip to main content
Version: Next

Color Space Conversion (RGB and CIELAB)

Files​

  • rgb2cielab.wgsl
  • cielab2rgb.wgsl

Overview​

Standard RGB (sRGB) is not perceptually uniform (see example below). CIELAB (L∗a∗b∗L^*a^*b^*) solves this, making it critical for accurate Bilateral Filtering and K-Means clustering.

  • L∗L^* (Lightness): 0.0 to 100.0
  • a∗a^* (Green-Red): Negative to Positive floats.
  • b∗b^* (Blue-Yellow): Negative to Positive floats.

:::info Example A mathematical difference of the same magnitude looks stronger in green than in blue.

Top Row (left-to-right):

  • red = rgb(100, 0, 0)
  • green = rgb(0, 100, 0)
  • blue = rgb(0, 0, 100)

Bottom Row (left-to-right):

  • red = rgb(150, 0, 0)
  • green = rgb(0, 150, 0)
  • blue = rgb(0, 0, 150)

:::

Implementation Details​

Because LAB requires negative numbers and values exceeding 1.0, the output texture must be a 32-bit float texture (texture_storage_2d<rgba32float, write>). If you attempted to store this in a standard 8-bit image (rgba8unorm), the data would clip between 0.0 and 1.0, destroying the LAB values.