PRTools contents |
IM_PATCH
B = IM_PATCH(A,PSIZE,PNUM,TYPE)
B = IM_PATCH(A,PSIZE,COORD,'user')
W = IM_PATCH([],PSIZE,PNUM,TYPE)
B = A*W
Input | |
A | Dataset or datafile with (multi-band) object images dataset |
PSIZE | 2-dimensional patch size. If PSIZE is 1-dimensional square patches of size PSIZE x PSIZE are generated. In case PSIZE < 1, it is taken relatuive to the images size. In this case patches may become non-square for non-square images. Default 3 x 3 |
PNUM | Number of patches, see TYPE, default 1 |
COORD | Given set of N x 2 image coordinates of patch centra. This may be given either in pixels (ANY(COORD > 1) or relative to the image size (ALL(COORD <= 1). |
TYPE | 'syst': systematic sampling generating PNUM x PNUM patches 'rand': generation of a random set op PNUM patches 'user': user defined patch positions, see COORD Default: 'syst'. |
Output | |
W | Mapping performing the desired patch generation |
B | Resulting dataset or datafile with the same number of objects as in A. Single images are replaced by the patches. |
The object images (including their N bands) are sampled and windows of size PSIZE are generated, replacing the original image object. They are stored as [PSIZE(1) PSIZE(2) NUM*N] image objects, in which N is the original number of bands and NUM is either PNUM (for TYPE is 'rand' or 'user') or PNUM x PNUM (for TYPE is 'syst').
By BAND2OBJ(B,N) individual patches can be transformed into objects.
B = IM_PATCH(A,0.5) % generate just the central parts of the images
B = IM_PATCH(A,[3 5],2) % generates 4 patches of size 3x5 in centra of
% the four quadrants
B = IM_PATCH(A,0.1,10,'rand') % generate at random positions 10 patches
%each with a linear size of 0.1 of the image size.
datasets, datafiles, band2obj,
PRTools contents |