Creates multiple, slightly jittered versions of an image.
Takes an image I, and generates a number of images that are copies of the
original image with slight translation, rotation and scaling applied. If
the input image is actually an MxNxK stack of images then applies op to
each image. Rotations and translations are specified by giving a range
and a max value for each. For example, if mPhi=10 and nPhi=5, then the
actual rotations applied are linspace(-mPhi,mPhi,nPhi)=[-10 -5 0 5 10].
Likewise if mTrn=3 and nTrn=3 then the translations are [-3 0 3]. Each
tran is applied in the x direction as well as the y direction. Each
combination of rotation, tran in x, tran in y and scale is used (for
example phi=5, transx=-3, transy=0), so the total number of images
generated is R=nTrn*nTrn*nPhi*nScl. Finally, jsiz controls the size of
the cropped images. If jsiz gives a size that's sufficiently smaller than
I then all data in the the final set will come from I. Otherwise, I must
be padded first (by calling padarray with the 'replicate' option).
USAGE
function IJ = jitterImage( I, varargin )
INPUTS
I - image (MxN) or set of K images (MxNxK)
varargin - additional params (struct or name/value pairs)
.maxn - [inf] maximum jitters to generate (prior to flip)
.nPhi - [0] number of rotations
.mPhi - [0] max value for rotation
.nTrn - [0] number of translations
.mTrn - [0] max value for translation
.flip - [0] if true then also adds reflection of each image
.jsiz - [] Final size of each image in IJ
.scls - [1 1] nScl x 2 array of vert/horiz scalings
.method - ['linear'] interpolation method for imtransform2
.hasChn - [0] if true I is MxNxC or MxNxCxK
OUTPUTS
IJ - MxNxKxR or MxNxCxKxR set of images, R=(nTrn^2*nPhi*nScl)
EXAMPLE
load trees; I=imresize(ind2gray(X,map),[41 41]); clear X caption map
% creates 10 (of 7^2*2) images of slight trans
IJ = jitterImage(I,'nTrn',7,'mTrn',3,'maxn',10); montage2(IJ)
% creates 5 images of slight rotations w reflection
IJ = jitterImage(I,'nPhi',5,'mPhi',25,'flip',1); montage2(IJ)
% creates 45 images of both rot and slight trans
IJ = jitterImage(I,'nPhi',5,'mPhi',10,'nTrn',3,'mTrn',2); montage2(IJ)
% additionally create multiple scaled versions
IJ = jitterImage(I,'scls',[1 1; 2 1; 1 2; 2 2]); montage2(IJ)
% example on color image (5 images of slight rotations)
I=imResample(imread('peppers.png'),[100,100]);
IJ=jitterImage(I,'nPhi',5,'mPhi',25,'hasChn',1);
montage2(uint8(IJ),{'hasChn',1})
See also imtransform2
Piotr's Computer Vision Matlab Toolbox Version 2.65
Copyright 2014 Piotr Dollar. [pdollar-at-gmail.com]
Licensed under the Simplified BSD License [see external/bsd.txt]