2d multi-peak detector (matrix)

[tinnitus]

i see, there was missunderstood with 20(3;2) and 20(3;5)

you don't look for a max but for high point comparing to montain's relief  ....

i search ................

[tinnitus]

the thing i see is to scan your matrix whatever the size with 3*3 matrix and seeing
if every point around is inferior, by scanning like your will find all peaks or sub peaks
after then you recover every peaks  you  have just to detect witch is the higher.

i 'm trying to write it, in labview, but i'm not still expert  so it take  time...

in literrature you found, what is method they used ?



 
 

[Marco_Ge]

Yes yours seems to be a good idea... it is quite similar to convolution operations in image processing.

I had no immediate results for my task in my researches, but i thought to use some approach with part derivatives.
This is surely most complex and maybe not friendly as the method you suggested.

I will try to build something...and let you know

Bye

marco

[Marco_Ge]

I think I have found something... in the annex I modified an example found over Convolution operations.

With a laplacian operator mask 5x5 it seems to work.

What do you think about it?

have a nice day,

Marco

[Marco_Ge]

sorry the exact mask is in this .vi

[tinnitus]

hi,

the principle is here but i don't think convoluate it's a good idea

because it look for a shape who look like and you could be lead in error

like this (see file and jpg) :      real max is -1 (1,1) and it found -2 (3,4)

more i don't think a 5*5 scan is good because that merge too much point  ...


i don't found formula but try to take a look with topography calculus....

if not,  method i suggested you is unstoppable even if it's a little bit nag

Best

 

[tinnitus]

Hello ,

are You still interested in ?

[Marco_Ge]

Hi and sorry for my delay... I was out of office and I returned only a few days ago.

I have tested and confirmed that the convolution method works fine with various real sets of data.
Results for my application seems to be good if I use the following convolution matrix (3X3):

0  -1  0
-1  10  -1
0  -1  0

I've only expanded my start data matrix with a border containing a very low value (-200) before the convolution operation, as real values span from -90 to -50 dB.

After the convolution operation, it is possible to locate exactly the maximum local peak . I have extracted these indexes to point out the real value in the starting matrix of data.

Thank you a lot for your helpful suggestions!

Bye and have a good day!

Marco

[tinnitus]

hello,

i made some progress since my beginings. As its awakes my curiousity i

finally approach a sympatic result without entries manipulations, the algo can still be improve, border thikness..., just tell me...

I 'll be interested to see behaviour and speed with very large array if you a one in one hand !

with this one i put a waiting time just to visualize progression , it can be removed

Best regards

Sébastien

[Marco_Ge]

Hi, first thank you for your tests and your interest. I am sorry I had a month out of office and so I could not make tests.
I checked out your vi with real values... here is attached with them.
But maybe something is working not too fine...
Actually I am using the approach I described some posts ago and it seems to work fine.
Thank you very much !
Marco