2D Segmentation

The class Seg2D provides a tool for automatic segmentation of two-dimensional multi-variate (multi-band) images.

Description: This class performs segmentation of an image into color-homogeneous segments. The homogeneity or heterogeneity level is specified as sigma. The three edit boxes denoted as minsigma, maxsigma and deltasigma denoted the homoegeneity level used when the image segmented first time (minsigma), last time (maxsigma) and with sigmas that are incremented by deltasigma (see the dialog below).

>

The segmentation is hierarchical in its nature and the exit criterion can be specified either by the minimum number of found segments (flagN2FindS is checked and the number of found segments is less than N2FindS in the corresponding edit box then exit) and

or by the maximum sigma. The former option is enforced by checking the flagN2FindS check box. Using this option, an image can segmented as illustrated below.

Input image.

Segmented image.

There is also an option to create a stack of segmentations by checking the check box labeled as flagN2FindLayers. With this option the range of maxsigma - minsigma will be divided into equal parts, where the number of parst is specified in the edit box N2FindLayers. The resulting segmentation will contain N2FindLayers-deep segmentation hierarchy that represents the intermediate segmentation results from minsigma to maxsigma.

Author: Peter Bajcsy, Y-J. Lee
Documentation: Peter Bajcsy.