OpenCV HSV Color Mask & Range Finder

Understanding the OpenCV HSV Color Space

In standard computer graphics, digital images are represented in the RGB (Red, Green, Blue) color model. While intuitive for display hardware, the RGB color space is notoriously brittle for computer vision tasks. In RGB, color chromaticity and luminance (brightness) are intertwined across all three channels. When a scene undergoes a slight lighting change or shadow cast, the $R$, $G$, and $B$ values of a target object shift unpredictably, causing rigid thresholding algorithms to fail completely.

The HSV (Hue, Saturation, Value) cylindrical representation decouples color information into three independent channels:

• Hue ($H \in [0, 179]$ in OpenCV): Represents the pure dominant color angle on a color circle ($0^{\circ}$ to $360^{\circ}$ scaled by half to fit within an 8-bit unsigned integer uint8).
• Saturation ($S \in [0, 255]$): Measures color purity and intensity (0 represents washed-out grayscale, 255 represents vibrant pure color).
• Value ($V \in [0, 255]$): Measures brightness and luminance (0 represents complete darkness, 255 represents maximum brightness).

Comparison of Color Spaces for Object Segmentation

To understand why HSV is preferred over RGB or Grayscale for object detection:

Why Hue Wraps Around the Red Spectrum in OpenCV

One of the most frequent pitfalls in color segmentation involves detecting red objects (such as the red game tiles in Okey Helper). Because red is located at $0^{\circ}$ ($360^{\circ}$) on the cylindrical hue wheel, red values wrap around the boundary:

• Lower Red Spectrum: $H \in [0, 10]$
• Upper Red Spectrum: $H \in [170, 179]$

To segment red objects comprehensively without false negatives, two separate binary masks must be computed and merged using bitwise OR operations:

import cv2
import numpy as np

# Dual-mask technique for circular hue wrap-around
lower_red1 = np.array([0, 120, 70])
upper_red1 = np.array([10, 255, 255])
lower_red2 = np.array([170, 120, 70])
upper_red2 = np.array([179, 255, 255])

mask1 = cv2.inRange(hsv, lower_red1, upper_red1)
mask2 = cv2.inRange(hsv, lower_red2, upper_red2)
final_red_mask = cv2.bitwise_or(mask1, mask2)

Recommended Calibration Workflow for Real-World Vision Pipelines

When tuning HSV ranges for industrial automation or game screen monitoring:

1. Isolate Dominant Hue: Set Saturation ($S_{\text{min}}=50, S_{\text{max}}=255$) and Value ($V_{\text{min}}=50, V_{\text{max}}=255$) broadly. Narrow the Hue slider range until only the target object's chromaticity is isolated.
2. Adjust Saturation Floor: Increase $S_{\text{min}}$ to filter out grayish background clutter, white UI borders, and ambient highlights.
3. Adjust Value Floor: Adjust $V_{\text{min}}$ and $V_{\text{max}}$ to ensure shadows do not introduce false positive voids while bright reflections are tolerated.
4. Apply Morphological Closing: In Python, pass the resulting mask through cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel) to fill internal holes before calculating contour moments.

Post-Processing: Contour Detection & Bounding Box Extraction

Once a clean binary mask is obtained using calibrated HSV bounds, the next step in a complete vision pipeline is extracting geometric contour hierarchies using cv2.findContours:

import cv2

# Find external contours
contours, _ = cv2.findContours(final_red_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

for cnt in contours:
    area = cv2.contourArea(cnt)
    # Filter noise blobs below minimum pixel area
    if area > 450:
        x, y, w, h = cv2.boundingRect(cnt)
        aspect_ratio = float(w) / h
        # Validate rectangular aspect ratio
        if 0.6 < aspect_ratio < 1.4:
            cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 2)