Image background in image recognition

Hi!

I want to find all occurences of this symbol in an image and is currently using CORR2 with a function that steps pixel-by-pixel through a larger image and try to find this symbol. The problem i find is that there are multiples of this symbol partially overlapping which it doesnt recognize as the image i compare to has a white background. How can I make corr2 (or a similar setup) to only match based on the black arrms of the symbol and not the background?

(Although image is attached it does not show, it is essentially a symbol with two crosses overlayed with one tilted 45 degrees.)

[Mahesh Taparia]

Hi,

Can you upload your image and specify the symbol you want to find?

[Mats]

Here you go!

[Image Analyst]

Yep, I think my code (in my answer below) will find it. Did you even try it? If not, why not?

[Mats]

Hi!

Haven't tried it due to winter holidays =)

Will try it this week and get back to you afterward

Answer 1

[Image Analyst]

See attached demo that uses normxcorr2() to find a template in a larger image.

[Mats]

Hi, I tried that one in 16 iterations where in each iteration I removed the mark it had identified and rerun the function. The result starts out good but then diverges and finds all sorts of other symbols (see attached screenshot).

Do you have a suggestion how to enforce the oblique lines in the correlation?

[Image Analyst]

Try raising the threshold to make sure it gets regions that are only really VERY close to the template.