R=VL_MSER(I) computes the Maximally Stable Extremal Regions (MSER)  of image I with stability threshold DELTA. I is any array of class UINT8. R is a vector of region seeds.

A (maximally stable) extremal region is just a connected component of one of the level sets of the image I. An extremal region can be recovered from a seed X as the connected component of the level set {Y: I(Y) <= I(X)} which contains the pixel o index X.

The function supports images of arbitrary dimension D.

[R,F]=VL_MSER(...) also returns ellipsoids F fitted to the regions. Each column of F describes an ellipsoid; F(1:D,i) is the center of the elliposid and F(D:end,i) are the independent elements of the co-variance matrix of the ellipsoid.

Ellipsoids are computed according to the same reference frame of I seen as a matrix. This means that the first coordinate spans the first dimension of I.

Notice that for 2-D images usually the opposite convention is used (i.e. the first coordinate is the x-axis, which corresponds to the column index). Thus, if the function VL_PLOTFRAME() is used to plot the ellipses, the frames F should be `transposed' as in F = F([2 1 5 4 3],:). VL_ERTR() exists for this purpose.

VL_MSER(I,'Option'[,Value]...) accepts the following options

Delta 

Set the DELTA parameter of the VL_MSER algorithm. Roughly speaking, the stability of a region is the relative variation of the region area when the intensity is changed of +/- Delta/2.

MaxArea [0.75]

Set the maximum area (volume) of the regions relative to the image domain area (volume).

MinArea [3 / numPixels]

Set the minimum area (volume) of the regions relative to the image domain area (volume).

MaxVariation [0.25]

Set the maximum variation (absolute stability score) of the regions.

MinDiversity [0.2]

Set the minimum diversity of the region. When the relative area variation of two nested regions is below this threshold, then only the most stable one is selected.

BrightOnDark 

Detect bright-on-dark MSERs. This corresponds to MSERs of the inverted image.

DarkOnBright 

Detect dark-on-bright MSERs. This corresponds to MSERs of the original image.

Verbose

Be verbose.

REFERENCES

 J. Matas, O. Chum, M. Urban, and T. Pajdla, "Robust wide baseline stereo from maximally stable extremal regions," in Proc. BMVC, 2002.