grad_vol = 0
filt_method = Pixel_Sigma
min_basin_size = 9
sigma_thresh = 30
iterations = 1
min_neighbors = 50
This filtering function operates as follows. Input to the process are the grayscale image and a gradient magnitude image. Another function, such as a Sobel or morphological gradient operator, can be used to compute the gradient image.
Based on the gradient image, a watershed analysis step is performed. This step divides the image into different locally uniform areas called catchment basins. Once this representation is complete, the pixels that belong to each basin are averaged to generate a single grayscale intensity for each basin, then filtering is performed in the catchment basin domain. Once filtering is complete, an output image is generated based on the final filtered basin results.
There are four different filtering methods that can be employed with this function through selecting the different options in the ``filt_method'' parameter selection. The ``Basin_Average'' filter is the weakest of the filters, which simply averages the intensities in a basin after watershed analysis and returns this as the filtered result. None of the other parameters are used with this filtering method. They are ignored.
The ``Pixel_Sigma'' option was used for testing purposes only and is not an effective filter. Do not use this option.
The ``Basin_FixedNeigh'' option defines the use of maximum homogeneity filtering in the basin domain based on a fixed set of neighbors. The top M% of the adjacent basins to the basin being filtered are used for local filtering, with this value being set with the ``Minimum neighbors to use (%)'' selection. This filter is generally not as effective as the basin-based sigma filter, below. The other options in the parameter menu are not used by this filtering method.
The ``Basin_Sigma_Filter'' option is the default, and most effective, of the filters. This option determines when neighboring basins are within T grayscale intensity values of the current basin, with T being set with the ``grayscale sigma threshold T'' option. If there is insufficient support for a basin based on a given value of T, then the filter drops back into a filtering method which is similar to the ``Basin_FixedNeigh'' option, and uses the top M% of the neighbors that match the current basin, as established by the ``min_neighbors'' option. This filter can also be iterated with the ``iterations'' selection.
The ``Basin_Sigma_Filter'' is the recommended filter for best results. The default values should be used except for the value of the ``Grayscale sigma threshold T.'' To set the value for T, it is best to sample a uniform region of the image to be analyzed, and use the standard deviation of the region as the value of T. Usually, 5 iterations of filtering will yield very smooth results.
The ``min_basin_size'' parameter sets the smallest catchment basin that can be generated by the watershed analysis. This value is recommended to be 9 pixels for 2D images (3x3) and 27 voxels for 3D imagery (3x3x3).