try this one

bw = pic > imdilate(pic, [1 1 1; 1 0 1; 1 1 1]);

or this one

CC = bwconncomp(BW);

for i = 1:CC.NumObjects,

index = CC.PixelIdxList{i};

if (numel(index) > 1),

BW(index) = false;

end

end

There is no specific command for that

The LocalMaximaFinder object finds local maxima in matrices.

After constructing the object and optionally setting properties, use the step method to find the coordinates of the local maxima in the input image. Use the step syntax below with input matrix, I, LocalMaximaFinder object, H, and any optional properties.

IDX = step(H,I) returns [x y] coordinates of the local maxima in an M-by-2 matrix, IDX. M represents the number of local maximas found. The maximum value of M may not exceed the value set in the MaximumNumLocalMaxima property.

[...] = step(H,I,THRESH) finds the local maxima in input image I, using the threshold value THRESH, when you set the ThresholdSource property to Input port.

I will try them and will report what happened. Thank you very much!

Wavelet analysis can be a good tool, especially because it can also detect the "shape" of your local maximum, i.e. its singularity, which is given by its Hölder exponent.

In wavelet multifractal analysis, the search for a specific local maximum is done with a wavelet whose number of moments is chosen in order to remove the regularity part of the signal. Thus the singularity part only remains. So back to your question, this means that, locally, you also can choose the "strength of the singularity" to detect the "type" of local maximum. Wavelets like derivatives of a Gaussian (first, second or Mexican hat) are good wavelets for detection of local maxima. And the power of wavelet in position detection makes it able to chain maxima through scales. So depending if your maxima are slowly varying signals of very peaky ones, you will be able to make the distinction between them.

You can simply use imregionalmax matlab function

Are you interested in finding the maximum intensity point in a specified neighbourhood?

Thanks to all of you for your answers. I think I should be more specific about the issue. The program I developed shows a radar image with different number of scatterers (or radar contacts, if you prefer) and the user must select which one is the interesting one. When doing so, the program shows the profile in both axes.

The problem that arises was that an user can not select the maximum value of the pixel, because our eye hasn't got that precision in differentiating between close pixels. My intention was to look for the maximum in the vicinity of the selected point.

Now, I have a prototype that finds the maximum in a matrix around the specified point. However, it is not working properly yet and I don't know how. So I will try some of your suggestions.

@Patrice Brault: could you give me some link or any source about the wavelet analysis? I think that it'll be interesting, maybe a little complicated, but interesting anyway.

@Pedro Guimaraes: (sorry, but I don't know how to put ~ above a letter). I will try that function and I will report what happend.

Thanks to all.

Hi Victor,

check this :

For wavelet analysis, the OWT (or DWT) Orthogonal Wavelet Transform will find the contours (or singularities), i.e. the pixels with a strong gradient, that correspond to a region of maxima. When the image variation is slow, the wavelet will find maxima at a low analysis scale. Check any page showing the famous Lena decomposed on an ortho. wavelets basis.

Finding a specific strength of a singularity, and chaining them through scales, can be done with a CWT (continuous WT, with no decimation between scales).

For this last analysis, check the papers of Mallat and Hwang, Mallat and Zhong, but also the Donoho-Johnstone toolbox (Wavelab), and the book of Alain Arneodo "Wavelets, multifractals and turbulences from dna to crystal growth" ,1997

You may also want to see this page with simpler Matlab functions (IMREGIONALMAX and BWCONNCOMP) : http://stackoverflow.com/questions/1856197/how-can-i-find-local-maxima-in-an-image-in-matlab (answer already given by Pedro Guimarães above; for people who know Latex, you can write \tilde{a} ;) )

LATEX! How I didn't realize it?! Thank you very much for your information.

Hi, I think your best option is a simple an easy gradient search. x0= clicked ponit; calculate numerically the gradient on x0, move one pixel in that direction, that pixel becomes the new x0, repeat until gradient is null.

I didn't ask finally but it was because I changed my job. I'm not working on this project anymore so I finally didn't implement all the solutions you provided me. Anyway, I would like to say thank you to everyone and I will try to get some results, however they won't be useful to my actual job :)

But the results may help somebody else. Thank you again.

Try the followings in MATLAB:

BW = imregionalmax(I);

BW = imregionalmin(I);

Use the LocalMaximaFinder object in the Computer Vision System Toolbox. It have worked just fine for me.

hLocalMax = vision.LocalMaximaFinder;    % create the object

hLocalMax.MaximumNumLocalMaxima = n;      % amount of maximal point

hLocalMax.NeighborhoodSize = [m m];          % Neighborhood Size

hLocalMax.Threshold = th;                % Threshold

location = step(hLocalMax, I) %Apply the object to the Image I. location is an nx2                                                                 %matrix  containing the coordinates of the maxima in                                                             %the image