Post

Statistical Measures Commonly Used in Computer Vision

Statistical Measures Commonly Used in Computer Vision

📊 Why Statistics Matter in Computer Vision

Computer vision systems rely heavily on statistics to:

  • Handle noise and uncertainty
  • Make decisions from pixel distributions
  • Design robust algorithms

Most classical CV algorithms are statistical at their core.


📐 Mean (Average)

Definition

\(\mu = \frac{1}{N} \sum_{i=1}^{N} x_i\)

Usage

  • Global / local brightness
  • Adaptive thresholding
  • Integral image based methods

Characteristics

  • Simple and fast
  • Sensitive to outliers

📉 Variance

Definition

\(\sigma^2 = \frac{1}{N} \sum_{i=1}^{N} (x_i - \mu)^2\)

Usage

  • Texture strength
  • Focus measure
  • Otsu thresholding

Characteristics

  • Measures spread
  • Noise sensitive

📏 Standard Deviation

Definition

\(\sigma = \sqrt{\sigma^2}\)

Usage

  • Adaptive thresholding (Niblack, Sauvola)
  • Noise estimation

Characteristics

  • Same unit as data
  • Intuitive interpretation

📊 Histogram

Definition

Frequency distribution of pixel intensities.

Usage

  • Global thresholding
  • Histogram equalization
  • Otsu / Entropy methods

Characteristics

  • Captures global distribution
  • Loses spatial information

🧠 Entropy

Definition

\(H = - \sum_i p(i) \log p(i)\)

Usage

  • Maximum entropy thresholding
  • Texture complexity
  • Information-based segmentation

Characteristics

  • Measures uncertainty
  • Robust to illumination changes

🔗 Covariance

Definition

\(\text{Cov}(X,Y) = E[(X-\mu_X)(Y-\mu_Y)]\)

Usage

  • PCA
  • Feature correlation
  • Motion analysis

Characteristics

  • Directional relationship
  • Scale dependent

🔄 Correlation Coefficient

Definition

\(r = \frac{\text{Cov}(X,Y)}{\sigma_X \sigma_Y}\)

Usage

  • Template matching
  • Stereo correspondence

Characteristics

  • Normalized
  • Range [-1, 1]

📌 Median

Definition

Middle value of sorted data.

Usage

  • Median filtering
  • Background modeling

Characteristics

  • Robust to outliers
  • Slower than mean

🧮 Percentile / Quantile

Definition

Value below which a percentage of data falls.

Usage

  • Contrast stretching
  • Robust thresholding

Characteristics

  • Resistant to noise
  • Distribution-aware

⚖️ Skewness

Definition

\(\text{Skewness} = E\left[ \left( \frac{x-\mu}{\sigma} \right)^3 \right]\)

Usage

  • Histogram shape analysis
  • Illumination bias detection

Characteristics

  • Detects asymmetry
  • Sensitive to noise

📈 Kurtosis

Definition

\(\text{Kurtosis} = E\left[ \left( \frac{x-\mu}{\sigma} \right)^4 \right]\)

Usage

  • Texture classification
  • Outlier detection

Characteristics

  • Measures tail heaviness
  • Sensitive to extreme values

🛡️ Robust Statistics (CV Practice)

StatisticRobust to OutliersTypical Use
MeanFast estimation
MedianNoise removal
MADRobust spread
PercentileContrast control

🧠 Practical Insight

Typical CV pipeline:

1
2
3
4
5
6
7
Image
 ↓
Local / Global Statistics
 ↓
Decision (Threshold / Model)
 ↓
Geometry / Logic

Understanding which statistic to use is often more important than the algorithm itself.


🎯 Takeaway

Statistics are the decision engine of computer vision.

Choosing the right measure:

  • Improves robustness
  • Reduces parameter tuning
  • Explains algorithm behavior

Classical CV = Statistics + Geometry + Logic 🚀

This post is licensed under CC BY 4.0 by the author.