Machine Vision

cancel
Showing results for 
Search instead for 
Did you mean: 

how to register a stained tissue image and an unstained tissue image

Hi folks,

 

I am currentlty doing image registration with a stained image and an unstained one, but I am not able to use any algorithms of the pattern matching to register these two images. Are the algorithms in pattern matching not sufficient enough to perform the registration of two distorted images like those in the attachments? Appreciate for any help.

stained.jpgunstained.jpg

0 Kudos
Message 1 of 10
(4,275 Views)

Please correct me if I am wrong, stained and unstained tissues differs in color.

 

If so you can use the pattern matching for color recognition and not for the pattern on the tissues because pattern matching algorithm needs to have a specific region on which you want to match the score between two images.

 

You can simply do the single color plane extraction for the image and then threshold it to analyse the particles and compare between samples.

 

If this is not something that you are expecting please elaborate your query.

0 Kudos
Message 2 of 10
(4,270 Views)

Thanks for your help. The pattern matching automaticale converes the image to grayscale so I surpose difference in color is not an issue. I have selected the ROI to search for similar parttern and the algorithm works very well with two indentical but rotated and scaled images, but it still cannot work with the stainned and unstainned images.

0 Kudos
Message 3 of 10
(4,264 Views)
May I know which region you are checking for similar pattern?
Message 4 of 10
(4,261 Views)

I have seleted this region from the unstained image as my ROI 

roi1.jpg

Ideally the algorithm will match this pattern to the corresponding area of the stainned image which is 

roi2.jpg

 

They are both converted to the grayscale. The algorithoms work well with the images provided in the Labview example as below

IMG_1589.JPG

 

0 Kudos
Message 5 of 10
(4,254 Views)

hi,

 

I think you want to match that nucleus (white region),correct?

 

Why are you preferring pattern matching only? Please go through various examples. There are far better methods than pattern matching.

 

Note: In your case it is possible to trace the nucleus but you have to add some preprocessing algorithm(like filters/grey morphology etc.)on template as well as on test image.

The NI example you shared it is matching of pattern but of similar kind of template and test image. Your template (unstained) and test image(stained) both seems to be different

If you want to stick to the pattern matching you can select smaller region, saving different templates for match otherwise I would recommend to go with the particle analysis with the thresholding approach.

 

Please let us know.

 

0 Kudos
Message 6 of 10
(4,237 Views)
the thing that you need is not pattern matching
it is sufficient geometry matching but before using this you have to use some thereshoudling and morphology code lie removing small abject
but if you insist in using pattern matching fist normalize your images then use it in this case stained will not bother you so much
0 Kudos
Message 7 of 10
(4,231 Views)

Hi,

 

Thank you very much for your advices. I think I have got some progresses on it through the threshholding and partical analysis.  The pattern macthing succeed in finding the rotation of two images.

IMG_1597.JPG

 

does it look reasonable to you?

 

Many Thanks.

0 Kudos
Message 8 of 10
(4,179 Views)

Hi,

Thanks a lot for your help. Here is what I got with pattern matching after removing small objects,

IMG_1597.JPG

Is it correct to what you suggested?

 

Thanks.

0 Kudos
Message 9 of 10
(4,178 Views)
also try to normalize or equalize your images before start pattern matching
0 Kudos
Message 10 of 10
(4,167 Views)