Feature extraction is the operation to extract various image features for identifying or interpreting meaningful physical objects from images.
Features are classified into three types.
Spectral features: color, tone, ratio, spectral index etc. Principle components and normalized vegetation index are widely used.
The first and second principle components computed from satellite multi-spectral scanner data such as Landsat TM will give "brightness" and "greenness" respectively.
Normalized difference vegetation index (NDVI) is often used for vegetation classification.
Generally spatial filtering (see 5-5) is applied as follows.
Step 1: smoothing filter such as mean or median is applied to avoid high frequency noises.
Step 2: edge detection filter such as Sobel, Laplacian or Highpass is applied to detect edges.
Step 3: line edges are detected by thinning and sometimes edge closing.
Textural features: pattern, homogeneity, spatial frequency, etc.
Though many computer approaches have been tried, human pattern recognition is much better than the computer results.
Table 5.3 summarizes major operations of feature extraction.