VL_PHOW Documentation

Question

The documentation for VL_PHOW on the two below websites is really unclear:

http://www.vlfeat.org/api/sift.html

http://www.vlfeat.org/matlab/vl_phow.html

Can someone explain to me how I can use VL_PHOW to access every pixel in an image. I want to be able to get the associated histogram for every pixel.

However please explain what the parameters are because I will want to edit that afterwards. I want to get the same number of features by using this technique and HOG and LBP.

IPath = fullfile(vl_root, 'data', 'roofs1.jpg') ;
I = imread(IPath) ;
I = im2single(rgb2gray(I)) ;
size(I)

[frames_dsift, descrs_dsift] = vl_phow(I, 'Sizes', 1, 'WindowSize',1 ,'Step', 1, 'Magnif', 6) ;
%[frames_dsift, descrs_dsift] = vl_phow(I) ;
size(frames_dsift)


[featureVector_hog, hogVisualization] = extractHOGFeatures(I);
%size(featureVector_hog)
%figure;
%imshow(I); hold on;
%plot(hogVisualization);


features_lbp = vl_lbp(I, 1);
%size(features_lbp)

Answer 1

As indicated on the links you have shared the PHOW computes DSIFT in several resolutions. So you have to understand how DSIFT works. Please see the following link. vl_dsift trying to get a feature vector at every pixel

It simply states that you cannot extract the features from every pixel. You have to skip the frame of 2 pixels thick from all four boundaries of the image (assuming your binsize is one). However, by using binsize of one, you can extract the features for all the pixels within that frame.

The explanation of the parameters you requested, can be found in the following link. http://www.vlfeat.org/matlab/vl_phow.html

Yours Sincerely...