Automatic multiresolution segmentation of three textures based on the red, green, and blue sample areas shown above. Using these samples, the Gabor filters were automatically designed and then used to generate the segmentation shown above.
Click here for larger image.
The teXan software package under development is a Sun Sparcstation, GUI-based interface for analysis of textured images. Features include:
To use Texan for this -> suggest constructing as large as possible of a 4-image composite image with upper&lower left quadrants being representative example tissues and with upper & lower right quadrants being unknowns. Then do biopsies of the upper two quadrants to train the filter-set. See if the resulting filter set properly segments the two images without biopsies in the other two quadrants. The whole thing runs automatically and generates the segmentation. 1. Select one biopsy as big as possible (red) for the known select a second (green) as big as possible (and preferably nearly the same general size and shape as shown in the example above) for unknown (use the edit drop-down menu to select biopsy pen color and size) 2. from the texture drop-down window select "Multi_filter...." 3. select: - initial prefilter sigma =2 (or 1) - Number of prefilter octave steps = 1 or 2 - segmentation error goal = 1 or 2 percent - post/pefilter sigma ratio = 1.5 ( 1.3 - 2 ) 4. filter parameters appear in the upper left window.