Computer Vision Color Spaces

A color space is a system for representing color as numbers. Different color spaces organize color information differently, and choosing the right one can make your computer vision task much easier.

Why Different Color Spaces Exist

Imagine you want to describe a person's skin tone under different lighting conditions. RGB values change dramatically as the lighting changes, even if the skin tone itself has not changed. Other color spaces separate brightness from color so that lighting changes affect fewer numbers — making detection more reliable.

The Problem with Raw RGB

Same orange under different lights:

Bright sunlight:  RGB = (255, 140, 0)
Dim room light:   RGB = (180, 100, 0)
Deep shade:       RGB = (100, 55,  0)

All three look "orange" to the human eye,
but the RGB numbers are completely different.
This makes it hard to reliably detect "orange."

The RGB Color Space

RGB stands for Red, Green, Blue. It is the most common color space in digital images. Each channel holds a value from 0 to 255. Mixing the three values creates the full spectrum of visible colors.

RGB works well for display purposes but has a weakness: it mixes brightness and color information together. This makes brightness changes affect all three channels at once.

RGB Color Cube

Think of RGB as a 3D cube:
         White (255,255,255)
        /|
       / |
      /  |
Blue /   | Green
    /    |
   /_____|
Black   Red (255,0,0)
(0,0,0)

Every color = a point inside this cube.

The HSV Color Space

HSV stands for Hue, Saturation, Value. It separates the color (hue) from how vivid it is (saturation) and how bright it is (value). This makes HSV ideal for detecting objects of a specific color regardless of lighting.

What Each Channel Means

HSV Cylinder Diagram

HSV is shaped like a cylinder:

         TOP (White/Bright, V=100%)
          ___________
         /     *     \       ← H=0° (Red) is at 3 o'clock
        |   *     *   |      ← H=120° (Green) is at 11 o'clock
        | *    +    * |      ← H=240° (Blue) is at 7 o'clock
        |   *     *   |      Center = zero saturation (gray)
         \___________/       Outer edge = full saturation
         BOTTOM (Black, V=0%)

To find orange: look at H ≈ 30°, high S, medium-high V.

Why HSV Is Better for Color Detection

Orange object detection goal: Find all orange pixels.

RGB approach:
  Check if R is between 200–255 AND G is between 100–180 AND B is 0–50.
  Under different lighting → range changes → detection fails.

HSV approach:
  Check if H is between 10°–25°.
  Lighting changes V (brightness) but NOT H.
  Detection stays reliable across lighting conditions.

The Grayscale Color Space

Grayscale removes all color and keeps only brightness. Every pixel gets one value between 0 (black) and 255 (white). Algorithms that do not need color information — like edge detection or shape analysis — work faster on grayscale images because there is one-third the data to process.

RGB to Grayscale Conversion

Formula (weighted average — eyes are most sensitive to green):
  Gray = 0.299 × R + 0.587 × G + 0.114 × B

Example pixel: RGB = (200, 150, 100)
  Gray = 0.299×200 + 0.587×150 + 0.114×100
       = 59.8 + 88.1 + 11.4
       = 159.3 → rounded to 159

Result: (159) — a medium gray

The LAB Color Space

LAB (also written L*a*b*) is designed to match how humans perceive color differences. It has three channels: L for lightness, A for green-to-red range, and B for blue-to-yellow range.

LAB Channels Explained

L channel: 0 (black) ───── 100 (white)
           Controls brightness only.

A channel: -128 (green) ── 0 (neutral) ── +127 (red)
           Controls green-red balance.

B channel: -128 (blue) ─── 0 (neutral) ── +127 (yellow)
           Controls blue-yellow balance.

LAB is useful when you want to measure how different two colors look to a human eye. A small numerical difference in LAB equals a small visual difference — this is not true in RGB.

The YCbCr Color Space

YCbCr separates luminance (Y, the brightness) from chrominance (Cb = blue difference, Cr = red difference). Video compression formats like JPEG and MPEG use this color space because the human eye is more sensitive to brightness than to color detail. Saving full resolution for Y and lower resolution for Cb and Cr cuts file size without visible quality loss.

YCbCr Use Case: Face Detection

Human skin spans many shades in RGB.
In YCbCr, skin tones cluster tightly:
  Cb: 77–127
  Cr: 133–173

This narrow range makes skin detection
much simpler and more reliable.

Choosing the Right Color Space

Key Takeaways

• A color space is a system for representing color as numbers.
• RGB is most common but mixes brightness and color together.
• HSV separates hue from brightness — ideal for color-based detection.
• Grayscale has one channel per pixel — faster processing, no color info.
• LAB matches human color perception — best for color difference measurements.
• Choose the color space that fits your specific computer vision task.