Remark. This implementation is considered legacy code and is superseded by VLFeat.
This is a MEX implementation of the Maximally Stable Extremal Region (MSER) feature detector [1]. The C source code consists of a single MEX file and should be very easy to port to environments other than MATLAB.
Overview
The package provides a single MEX function mser that
extracts the maximally stable extremal regions from a given
gray-scale image. So for example (this code is part of
the mser_demo2 M-file included in the package), lets
load a test image
>> I = load('clown') ; I = uint8(I.X) ;
>> figure ; imagesc(I) ; colormap gray; hold on ;
This should result in the following picture:
We run mser to extract regions r and
fitted ellipses ell:
>> [r,ell] = mser(I,uint8(5)) ; Sorting pixels ... done Computing extremal regions ... done Extremal regions: 3679 Maximally stable regions: 475 (12.9%) Bad regions: 0 Small regions: 432 Big regions: 15 Duplicated regions: 8 Cleaned-up regions: 20 (0.5%) Fitting ellipses... mean 0 mean 1 corr (0,0) corr (0,1) corr (1,1)
Regions are represented as point of maximum intensity of the corresponding connected component level set. Let us visualize these points:
>> [i,j]=ind2sub(size(I),r) ; >> plot(j,i,'r*') ;
This is not really meaningful. It is much more interesting to have
a look at the fitted ellipses (for this step you need the
function plotframe that can be found as part of
the VLUtil package):
>> ell = ell([2 1 5 4 3],:) ; >> plotframe(ell);
In order to see what a maximally stable extremal region looks
alike we need to compute a connected component of a level set. To
this end we use the function
erfill:
>> k = 1 ; % mser number >> clf ; >> sel = erfill(I,r(k)) ; >> mask = zeros(M,N) ; mask(sel) =1 ; >> imagesc(cat(3,I,255*uint8(mask),I)) ; colormap gray ; hold on ; >> plot(j(k),i(k),'r*') ; >> plotframe(ell(:,k),'color','r') ;
