A MATLAB library for efficiently generating Basic Image Features (BIFs) from images.

Type mtInstallBifLab at the Matlab command line to add BifLab folders to Matlab path

Type mtDemoBifLab at the Matlab command line to display demonstration images and their associated BIFs

Generate BIFs from an image using bifs = mtBifs(inputImage, blurWidth, flatnessThreshold)

blurWidth (σ in the reference papers) sets the scale of the features that the BIFs detect. It is the standard deviation of the gaussian derivative filters used to calculate the BIF responses .

flatnessThreshold (γ in the reference papers) is a threshold factor used to control the minimal response required for a non-flat BIF to be returned. This ensures that the flat BIF is selected where there is no strong response from any other BIF. The higher the flatness threshold factor, the more likely a pixel is to be considered flat. A value of 0.1-0.2 is a reasonable starting point. However, the best value will depend on the nature of the images used in each application so some experimentation is recommended.

Display BIFs using bifs.show(0) (BIF classes only) or bifs.show(1) (BIF classes and orientations). If no parameter is provided, the default is to show classes only.

It is recommended to show orientations only for small image areas (say up to about 50x50 pixels). For larger areas, it can be hard to distinguish the orientations and Matlab's image display features tend to get a bit slow.

bifs.getSnippet(rows, columns) can be used to easily get a snippet of a larger set of BIFs for display or further processing.

For further documentation of the BIF class, including its properties and methods, type doc mtBifs at the Matlab command line.

BIFs are a classification of local image structure into seven different classes on the basis of approximate local symmetry. The seven classes are:

1 = flat (pink). No orientation.

2 = gradient (grey). Orientation is "up" luminance gradient from dark to light.

3 = dark blob (black). No orientation.

4 = light blob (white). No orientation.

5 = dark line (blue). Orientation is along dark ridge.

6 = light line (yellow). Orientation is along light ridge.

7 = saddle (green). Orientation is along dark ridge.

Gradient, line and saddle BIFs also have an associated orientation, which is also computed for these features. Note that for line and saddle BIFs, orientations separated by 180 ° are equivalent. However, the library does not yet canonicalise these to a single orientation.

Crosier, M. & Griffin, L. D. (2010). Using Basic Image Features for Texture Classification. International Journal of Computer Vision, 88, 447-460, doi:10.1007/s11263-009-0315-0

Published paper from journal site | Author version from UCL's institutional repository.

Lillholm, M. & Griffin, L. D. (2008). Novel image feature alphabets for object recognition. 19th International Conference on Pattern Recognition (ICPR), 1-4, doi:10.1109/ICPR.2008.4761173

Published paper from journal site | Author version from UCL's CS department.

Run tests to verify the output of the library for canonical features for each BIF class using one of:

mtBifs_test(1); to print a test report in the Matlab command window.

mtBifs_test(2); to print test report and also show test images and BIFs for manual verification.

[testOutcomes, testOutcomeDetails] = mtBifs_test(0); to save the test outcomes and outcome messages for further processing with no printed output.

For more information on running the provided tests and defining additional test cases of your own, type doc mtBifs_test at the Matlab command line.

Run a test to verify the fast BIF histogram generation code produces the same BIF histograms as an idiot-proof pixel-by-pixel calculation using one of:

mtBifs_roiHistograms_test(1); to print a test report in the Matlab command window.

mtBifs_roiHistograms_test(2); to print test report and also show test image, BIFs and sample histogram for manual verification.

[testOutcomes, testOutcomeDetails] = mtBifs_roiHistograms_test(0); to save the test outcomes and outcome messages for further processing with no printed output.

O'Reilly, Martin (2016): BifLib Matlab BIF library. figshare. doi:10.6084/m9.figshare.3749217