how to know the number of regions of the image?

Hi! I have a color segmented image. i need to know the number of regions of the image , and the size of each region. Can anyone help me please?

[Ahmad Moniri]

Can you give an example image?

Answer 1

[Guillaume]

You use regionprops on your segmented image. The number of elements in the structure array it returns is the number of regions. The Area field or the BoundingBox field of the structure gives you the size, depending on what you call size.

[As Bdg]

Thank you for your answer ! I used watershed segmentation. But after the segmentation step, i did another processing for my image (inverse wavelet transform). Now i need to determine the regions of the new image (after segmentation and inverse wavelet transform)

[Guillaume]

I'm not sure what an inverse wavelet transform of a labeled image gives you. If it's still a label image then you can use regionprops on it. If it's not, then you need to segment it again using an appropriate method, possibly another watershed.

I don't really understand why you'd want to apply an inverse wavelet transform to the result of a watershed segmentation.

[As Bdg]

I'm using multiresolution approach for image segmentation. i constructed the pyramid using the wavelet transform , i segmented the image using watershed at a lower resolution, but now i need to go back to my original image . that's why i need to apply an inverse wavelet transform . And after applying it i have to merge the regions and refine regions borders, so i need to define the regions after the inverse wavelet transform. If i'll use another watershed, the image will be more over-segmented , however my goal is to avoid this over-segmentation and minimise the number of regions.

Answer 2

[Image Analyst]

Use bwlabel() on your segmented (binary) image - that's one way.

[~, numRegions] = bwlabel(segmentedImage);

[As Bdg]

But it is not a binary image. It's an RGB image. And when i applied the inverse wavelet transform, it is no more segmented. So i can't use bwlabel()

[Image Analyst]

To me, segmented means binary, or at the very least labeled, not a full color image. If you have a full color image with a limited number of colors, you can simply construct the 3-D histogram and count the number of non-zero bins, if each region of your image has a unique color.

[As Bdg]

In fact it's a full color image , and when i segmented it (with watershed algorithm) i got a labeled image.i used label2rgb and did the inverse wavelet transform , but i lost the regions which i need to merge. Sorry i don't understand well your idea. Would you please clarify more? I'll be thankfull

[Image Analyst]

Usually when people ask for image processing advice, they attach an image. However I don't have the wavelet toolbox so even if you had, all I could say is "Good luck" since I wouldn't be able to run your code.

[Guillaume]

I don't have the wavelet toolbox either, and I can't really picture in my head the result of an inverse wavelet transform on a segmented image. I don't even understand the goal here. You have transformed a segmented image into something else. If the image is no longer the segmented image what are you hoping to recover?

[As Bdg]

i'm using the multi-resolution approach for image segmentation.i applied the wavelet transform to construct the pyramid, then i segmented my image at level 3 (with watershed algorithm) but it is over segmented. I applied the inverse wavelet transform to go back in the pyramid. I want to merge the regions at the lower lever and recover regions borders. Here are two images : the first at level 3 after segmentation. The second is after inverse wavelet transform (level 2).

[As Bdg]

after inverse wavelet transform

[As Bdg]