License Information

Use of this function requires a license for Whitebox Workflows for Python Professional (WbW-Pro). Please visit www.whiteboxgeo.com to purchase a license.

Description

This tool is used to segment a mult-spectral image data set, or multi-dimensional data stack. The algorithm is based on region-growing operations. Each of the input images are transformed into standard scores prior to analysis. The total multi-dimensional distance between each pixel and its eight neighbours is measured, which then serves as a priority value for selecting potential seed pixels for the region-growing operations, with pixels exhibited the least difference with their neighbours more likely to serve as seeds. The region-growing operations initiate at seed pixels and grows outwards, connecting neighbouring pixels that have a multi-dimensional distance from the seed cell that is less than a threshold value. Thus, the region-growing operations attempt to identify contiguous, relatively homogeneous objects. The algorithm stratifies potential seed pixels into bands, based on their total difference with their eight neighbours. The user may control the size and number of these bands using the threshold and steps parameters respectively. Increasing the magnitude of the threshold parameter will result in fewer mapped objects and vice versa. All pixels that are not assigned to an object after the seeding-based region-growing operations are then clumped simply based on contiguity.

It is commonly the case that there will be a large number of very small-sized objects identified using this approach. The user may optionally specify that objects that are less than a minimum area (expressed in pixels) be eliminated from the final output raster. The min_area parameter must be an integer between 1 and 8. In cleaning small objects from the output, the pixels belonging to these smaller features are assigned to the most homogeneous neighbouring object.

The input rasters (inputs) may be bands of satellite imagery, or any other attribute, such as measures of texture, elevation, or other topographic derivatives, such as slope. If satellite imagery is used as inputs, it can be beneficial to pre-process the data with an edge-preserving low-pass filter, such as the bilateral_filter and edge_preserving_mean_filter tools.

See Also

bilateral_filter, edge_preserving_mean_filter

Project Links

WbW Homepage User Manual Support WbW