multi-scale Gaussian kernels

7 views (last 30 days)
Hi!
I'm working on the detection + classification of microaneurysms and have found many papers recommending a multiscale gaussian correlation coefficient. As I am new to image processing, I find the explanation in the original paper "Detection of microaneurysms using multi-scale correlation coefficients" by Zhang et al a bit complicated. Can anyone give me pointers on how to implement this?
I'm supposed to "apply a sliding neighborhood filter with multi-scale Gaussian kernels to the fundus image in order to calculate acorrelation coefficient for each pixel. Based on extensive experimentation, we chose five scales for the kernel to represent microaneurysms of different sizes with the sigma of the Gaussian function being 1.1, 1.2, 1.3, 1.4 and 1.5.The maximum coefficients from each of the five responses were combined to form a final response." <
>

Accepted Answer

Image Analyst
Image Analyst on 1 Jun 2017
Try imgaussfilt() and max().
  4 Comments
Stelmaria Lyra
Stelmaria Lyra on 1 Jun 2017
Ah yes thank you! I thought it was more complicated than that!
"The maximum coefficients from each of the five responses were combined to form a final response." implies that for each sigma there were multiple calculations?
From what I understood, I did this:
for sigma=1.1:0.1:1.5
Gauss = imgaussfilt(Green, sigma);
r = corr2(Gauss,Gray)
end
but it only gives one correlation coefficient for each sigma so i must be missing something?
Also, you suggested max(), what for exactly?
Image Analyst
Image Analyst on 2 Jun 2017
You have one original image, and 5 blurred images. I'm pretty sure you're not supposed to use corr2(). corrcoeff() gives a 2x2 array for the entire image. If you have megapixel images, you'd still get a 2x2 correlation matrix, not a megapixel image. They say you'd have 5 megapixel correlation coefficient images, and then (I guess), take the value from the correlation coefficient image that has the greatest value. I don't know how they get a correlation for every pixel in the image when comparing the original image with one blurred image. Sorry, you'll have to read the paper more carefully or ask the authors.

Sign in to comment.

More Answers (1)

Krishna Chaitanya
Krishna Chaitanya on 27 Oct 2019
Did you get the answer for this question,If so how did you implement it?@Stelmaria Lyra

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!