Merge Algorithms Background


Feature Extraction evaluates the spectral similarities of segments created from the watershed algorithm and merges them if their spectral properties are similar. First, it computes the mean spectral values for each band, in each adjacent segment. Then it computes the Euclidean distance between the segments using one of the following methods:

Full Lambda Schedule

This method iteratively merges adjacent segments based on a combination of spectral and spatial information. Merging occurs when the algorithm finds a pair of adjacent regions, i and j, such that the merging cost is less than a defined threshold lambda value:

Where:

is region i of the image

is the area of region i

is the average value in region i

is the average value in region j

is the Euclidean distance between the spectral values of regions i and j

is the length of the common boundary of and .

For more information on the Full Schedule Lambda method, see the following reference:

Robinson, D. J., N. J. Redding, and D. J. Crisp. Implementation of a fast algorithm for segmenting SAR imagery, Scientific and Technical Report. Australia: Defense Science and Technology Organization, 01 January 2002.

Fast Lambda

This method computes the Euclidean color distance (E) and the length of their common border (L). N1 and N2 are the numbers of pixels in regions 1 and 2, respectively:

Larger Lambda values represent:

  • Larger Euclidean color distances, OR
  • A smaller common border

Smaller Lambda values represent:

  • Smaller Euclidean color distances, OR
  • A larger common border

All adjacent regions with an L value less than a given threshold will be merged on an iterative basis. The threshold is computed from a normalized cumulative distribution function (CDF) of Lambda spectral distance values. For example, if you set the Merge Level to 20, it will merge adjacent segments with the lowest 20% of Lambda values.