SPATM

Augment image dataset with spatial label information

    E = SPATM(D,S)
    E = D*SPATM([],S)

Input
D	image dataset classified by a classifier
S	smoothing parameter (optional, default: sigma = 1.0)

Output
E	augmented dataset with additional spatial information

Description

If D = A*W*CLASSC, the output of a classification of a dataset A containing feature images, then E is an augmented version of D
E = [D T]. T contains the spatial information in D, such that  it adds for each class of which the objects in D are assigned to,  a Gaussian convoluted (std. dev s) 0/1 image with '1'-s on the  pixel positions (objects) of that class. T is normalized such that  its row sums are 1. It thereby effectively contains Parzen estimates  of the posterior class probabilities if the image is considered as a  feature space. Default: S = 1.

Spatial and feature information can be combined by feeding E into  a class combiner, e.g: A*W*CLASSC*SPATM([],2)*MAXC

See also

datasets, mappings, prex_spatm,