how can i delineate rgb images

Hi,

how can i delineate objects on rgb images , it seems that contour doesn't work for rgb images, here is an exmple of what i want

Thanks

Answer 1

[Image Analyst]

Looks like you'll have to do color segmentation. I have demos for that in my File Exchange. http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

Also check out published image analysis papers on how people found that thing. http://www.visionbib.com/bibliography/contentsmedical.html#Medical%20Applications,%20CAT,%20MRI,%20Ultrasound,%20Heart%20Models,%20Brain%20Models

[sbei arafet]

Hi,

thanks for the links i want to know why exemples that i have seen transform the rgb space to Lab or HSV and not working directly

regards

[Image Analyst]

Because often/usually RGB is not a good colorspace for color analysis or segmentation. For example if you wanted "green" things, you can't just threshold a single channel like you can with the hue channel. The green may happen at all levels of green channel, depending on what red and blue are doing. All that has to happen is that the green signal is higher than those other two. But now you're dealing with 3 channels and they can't even be thresholded by themselves. For example, look at the color gamut of your image in RGB color space:

Now there is no way you can threshold the 3 color channels to get the yellow pixels. Thresholding essentially carves out a rectangular box from that gamut. But any rectangular box in RGB space would capture some of those pink pixels also.

But what if we transform it to HSV color space:

As you look roughly down the V axis, you can see that if you threshold the Hue channel between about 30 and 90 degrees (or 0.08 and 0.25, you'll get all the yellow stuff. Some of the stuff near the center (the V axis) has low saturation, but you can threshold that out too. So you can get the yellow by doing something like this:

yellowPixels = hueChannel > 0.08 & hueChannel < 0.25 & sChannel > 0.3;

That will get all the vivid yellow pixels but not the white or red pixels.

[sbei arafet]

Hello thanks for the explanation

[Image Analyst]

So what's holding you up now?

[sbei arafet]

Hi,

thanks Image Analyst , in my case i can't decide that is the best way ,since i'm not segmenting a specific color , i have some informations about each modality for pet and mri images and i will applicate irfc (iterative relative fuzzy connectedness) algorithm to segment each layer, so the idea is :

1. 1- segment each layer with irfc
2. 2- extracting matching points for the 3 layers
3. 3- recuperate contour coordinates after finding intersected points
4. 4- than overlaying this contour on the color image

I don't know if it is the best way what i'm working on know is how to design the affinity function by analysing charecteristics of each modality , i will be grateful if you could answer this questions

• Giving SUV values for pet images and intensity for MRI , how could i analyse gaussian distribution of each modality , is it by applicating au gaussian mexture?
• how could i superposing x y coordinates from gray level on color images.

here is the paper i tried to contact the author but i can't " https://www.researchgate.net/profile/Ulas_Bagci/publication/282049691_Fuzzy_connectedness_image_co-segmentation_for_hybrid_PETMRI_and_PETCT_scans/links/560bfa2708ae73e7a6a2d207.pdf "

Thanks in advance