Merge regions

Hi, I want to merge two regions if they are close to each other and their average intensity is similar.

 if (| y1-y2 | <T & | M2-M1 |<t)

 statement to merge

 end

 But I do not know how to write the statement for the merger of regions. Anyone know?

[Walter Roberson]

How are the regions represented?

[Marlene]

regions are white in a binary image representing objects.

[Walter Roberson]

Suppose you had two squares separated by a modest distance, with one of the two a bit up and to the right of the other. What would you like the merged region to look like in that case?

Answer 1

[Image Analyst]

Something like this perhaps???:

if (meanOfRegion1 - meanOfRegion2) < maxAllowableDifference
  mergedMask = region1Mask & region2Mask; % Assumes binary images.
  mergedImage = originalImage; % Initialize new image.
  mergedImage(~mergedMask) = 0; % Zero out everything outside the merged region.
end

[Marlene]

yes, but the regions are in the same binary image. are regions that are next to each other, but can only be re merged if (meanOfRegion1 - meanOfRegion2) < maxAllowableDifference.

So, these steps resolved the problem?

[Image Analyst]

OK, but what do you think the mean of the binary regions is? I'll bet my next year's salary that they'll all be 1 (since logical true = 1). So what's the point? If all your regions are 1 (because they're from a binary image), then how can you decide whether to merge or not merge regions?

[Walter Roberson]

Presumably the selected areas mask an image which is not binary, and it is the average magnitude of the active parts of the second image which is being considered.

But still I don't know how Marlene would like the connection to be made between the two regions. Smallest box that encloses both regions? Just label both regions the same and let them continue to be disjoint? Figure out the closest approach of the two objects and create a 1-wide 8-connected path between the two ? My Magic 8-Ball says "The situation is clouded".

[Image Analyst]

I agree Walter, that's why my suggestion had both a binary image and an original grayscale image. In cases like these there is usually some sort of heirarchy that keeps track of what regions are adjacent to what other regions. Otherwise you can play tricks with binary or integer images to keep track. I don't know how she's keeping track. Adjacent regions are have the potential to be merged.

[Marlene]

Yes, I have a binary image with many separate regions, but some of these regions are part of the same object but is fragmented with segmentation. So I go to grayscale image and calculate the average intensity of each of these regions. Then I wanted to merge the regions that are close to each other and have similar average intensity.

[Image Analyst]

That's what we thought. So, yes, my code is essentially what you'd do and would "resolve the problem" if you choose to do it that way.

[Walter Roberson]

Okay, so given that you have located two nearby regions that you want to merge, how do you want to merge them?

For example, suppose that one of the regions was a square, and three pixels to the right of it was the closest point of a circle of the same diameter. What should the merged region look like?