Color Spaces

What Is a Color Space?

A color space is a coordinate system for representing colors numerically. Different color spaces are chosen depending on the application.

RGB Color Space

The RGB color space represents colors by additive mixing of three primary colors: Red, Green, and Blue. To treat it as a concrete color space, one specifies a standard such as sRGB that fixes the primaries, white point, and gamma (the same RGB values denote different colors under different standards).

RGB Cube

Additive Mixing

R, G, B ∈ [0, 255] (8-bit)   Total colors: 256³ ≈ 16.77 million

HSV/HSL Color Space

HSV (Hue, Saturation, Value) separates a color into hue, saturation, and value, which makes it intuitive to work with and well-suited for color selection and simple threshold-based extraction. It is not perceptually uniform, however (pure blue and white both have $V = 1$). HSL, also in the heading, is a variant that uses the average of the maximum and minimum, $L = (M + m)/2$ ($M, m$ are the maximum and minimum of R, G, B; the same quantity as the lightness method below), instead of the value $V$, and its saturation is defined differently.

Applications: Color-based object detection, skin detection, background subtraction, etc. For example, simple red detection restricts H to 0°–10° or 350°–360° and also sets lower bounds on S and V (adjust the ranges to the lighting and camera).

YUV / YCbCr Color Space

These color spaces separate a brightness signal Y (strictly speaking luma, not luminance) from chrominance. Since human vision is more sensitive to luminance than to color differences, the chrominance channels can be subsampled to reduce data size.

Lab Color Space (CIELAB)

Lab is a color space designed so that color differences are roughly proportional to perceived differences (approximately perceptually uniform). Euclidean distance in Lab roughly corresponds to perceived color difference, making it widely used for color difference evaluation and color management. It is not perfectly uniform, however, and the agreement varies across regions of color space.

Applications: Color difference calculation ($\Delta E$), color calibration, image segmentation. $\Delta E = \sqrt{(\Delta L^*)^2 + (\Delta a^*)^2 + (\Delta b^*)^2}$ serves as a simple perceptual color difference metric (the CIE76 color difference); CIE94 and CIEDE2000 ($\Delta E_{00}$) are used as metrics that agree better with perception.

Grayscale Conversion

There are several methods to convert a color image to grayscale (a single channel corresponding to brightness). Since the meaning of the value depends on the method, we write the result of the luminosity method as $Y$ and that of the other methods as $I$:

Code Example (Python)

import cv2
import numpy as np

# Read image (BGR)
img_bgr = cv2.imread('image.jpg')

# Color space conversions
img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
img_hsv = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2HSV)
img_ycrcb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2YCrCb)  # channel order: Y, Cr, Cb
img_gray = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2GRAY)
img_lab = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2LAB)  # 8-bit: stores L*×255/100 and a*, b* offset by 128

# Extract a specific color using HSV (e.g., red; OpenCV's H is 0–179, so 0°–10° → 0–5 and 350°–360° → 175–179)
mask1 = cv2.inRange(img_hsv, np.array([0, 100, 100]), np.array([5, 255, 255]))
mask2 = cv2.inRange(img_hsv, np.array([175, 100, 100]), np.array([179, 255, 255]))
mask = cv2.bitwise_or(mask1, mask2)
result = cv2.bitwise_and(img_bgr, img_bgr, mask=mask)

# Manual grayscale calculation (luminosity method, BT.601 weights; round to match cv2)
b, g, r = cv2.split(img_bgr)
gray_manual = np.round(0.299 * r + 0.587 * g + 0.114 * b).astype(np.uint8)

Frequently Asked Questions

Summary

• RGB: The standard for display, based on additive mixing
• HSV/HSL: Intuitive separation into hue, saturation, and value; well suited to color extraction
• YUV/YCbCr: Separates luma from chroma, used in video compression
• Lab: Approximately perceptually uniform, suited to color difference and color management
• Grayscale: A single channel corresponding to brightness (its meaning depends on the conversion method), widely used as preprocessing for image analysis