Indexing for partial matrix

I have a 140 x 2000 matrix. It contains a blob of area (I know the linear indices) that has known peaks. I want to find the maximum peaks in part of the blob. Specifically I want to find the maximum peak in rows 1-77 and the maximum in rows 78-140 of the blob. There may be other blobs with other peaks, so it has to be by blob area.

Should I generate the linear indices for each section (e.g. all columns, rows 1:77) and then find which are common with the blob indices? How would I do that?

Answer 1

[Matt J]

[upperBlob,lowerBlob]=deal(yourMatrix);

upperBlob(78:end,:)=-inf;
lowerBlob(1:77,:)=-inf;

[imax1,jmax1]=find(upperBlob==max(upperBlob(:)));  %upper blob max
[imax2,jmax2]=find(lowerBlob==max(lowerBlob(:)));  %lower blob max

[Stephen Thompson]

This doesn't quite do it. I needed the peaks within the boundary of the blob (yellow). I have since found the ind2sub function which helped be solve the problem.

[Matt J]

I don't see why it wouldn't have worked. The area outside your blobs appears to be zero, while the area inside the blobs appears to be greater than zero, so the maxima I've computed above would inherently have to fall within the blobs. Regardless, you could exclude the background with a few additional lines:

A=-inf(size(yourImage));
A(blobIndices)=yourImage(blobIndices):

[upperBlob,lowerBlob]=deal(A);

upperBlob(78:end,:)=-inf;
lowerBlob(1:77,:)=-inf;

[imax1,jmax1]=find(upperBlob==max(upperBlob(:)));  %upper blob max
[imax2,jmax2]=find(lowerBlob==max(lowerBlob(:)));  %lower blob max

[Stephen Thompson]

It is my fault for not explaining the problem well enough. Another example with 2 blobs - but I still would want them classed by row (corresponds for frequency but axes left generic) and by blob.

There could be multiple blobs with peaks above and below the row cut line, but I want to separate them into maxima by blob and by dichotomous row grouping, So it does have to be separated on a per-blob basis. Now that I can find the subscripts (ind2sub) for the blobs I can solve the problem. Thank you though for your solution.