Image Texture Features — Mean, GLCM Metrics, Entropy & Energy
Image Intensity Statistics and Texture Features
Mean (m)
Mean (m)
The mean is the expected value of the pixel intensity and reflects the overall brightness of the image.
↑ MEAN → ↑ BRIGHTNESS
Variance (M₂(Z))
Variance (M₂(Z))
The variance measures the dispersion of pixel intensities around the mean.
↑ VARIANCE → Large intensity differences and contrast textures, noise.
↓ VARIANCE → Smooth or uniform textures.
Standard Deviation (σ)
Standard Deviation (σ)
The standard deviation is the square root (√) of the variance, which expresses contrast in the same unit
of pixel intensity. It quantifies how much intensities deviate from the mean.
↑ STANDARD DEVIATION → Sharper transitions.
↓ STANDARD DEVIATION → Smooth surfaces.
Skewness
Skewness
The skewness is the normalized third moment and measures the asymmetry of the intensity distribution.
− SKEWNESS → The tail of the histogram goes toward ↑ intensities, dark pixels.
+ SKEWNESS → The tail of the histogram goes toward ↓ intensities, bright pixels.
Kurtosis
Kurtosis
The kurtosis is the normalized fourth moment and it quantifies the sharpness/flatness of the histogram
compared with a normal distribution.
↑ KURTOSIS → A narrow peak in which the intensities are near the mean.
↓ KURTOSIS → A flatter histogram where the intensities are spread out.
Entropy (intensity distribution)
Entropy
The entropy is the disorder of the intensity distribution.
↑ ENTROPY → It has many intensity values and complex textures.
↓ ENTROPY → It has uniform intensities, so repetitive textures.
0 ENTROPY → Constant image.
Energy (intensity distribution)
Energy
The energy is the uniformity of the intensity distribution, so it acts opposite to the entropy.
↑ ENERGY → The histogram is concentrated around a few values, indicating uniformity and smooth texture.
↓ ENERGY → The histogram has many values (many probability intensities), indicating noise, disorder,
and complex textures.
GLCM Parameters
- DISTANCE (d) → The displacement between two pixels being considered.
- ANGLE (θ) → The direction in which the pixel pairs are considered.
- NUMBER OF GRAY LEVELS (G) → The number of discrete intensity levels.
In the example: G = 4, θ = 0° and d = 1.
Contrast
Contrast
The contrast measures the local intensity variation between neighboring pixels by giving a weight to pairs
with large intensity differences.
↑ CONTRAST → Coarse textures with sharp changes.
↓ CONTRAST → Smooth and homogeneous regions.
Homogeneity
Homogeneity
The homogeneity quantifies how close intensity pairs are to the main diagonal of the GLCM.
↑ HOMOGENEITY → The neighboring pixels have similar intensities, which means smooth textures.
↓ HOMOGENEITY → There are large intensity differences.
Entropy (spatial distribution)
Entropy
The entropy is the disorder level in the spatial distribution.
↑ ENTROPY → Indicates a complex, irregular texture.
↓ ENTROPY → Corresponds to uniform and predictable patterns.
Energy (spatial distribution)
Energy
The energy measures the uniformity of repetition of the intensity pairs.
↑ ENERGY → Indicates regular texture with structured patterns and high probability of some intensity pairs.
↓ ENERGY → Indicates irregular or complex structures.
Correlation
Correlation
The correlation measures the linear dependency between the intensities of neighboring pixels.
↑ CORRELATION → The intensities are related in a predictable way.
↓ CORRELATION → The intensities have weak dependency.
Dissimilarity
Dissimilarity
The dissimilarity is the average intensity difference between neighboring pixels.
↑ DISSIMILARITY → There are more pronounced texture variations and roughness.
↓ DISSIMILARITY → There are smoother textures.