Quality of Turboreg output not sasisfying

[JoLi]

Hello everybody

First some information about my images, which are rgb images received via matlab. I transform them to 8 or 16 Bit grayscale images using imagej. I have a slight drift and rotation of my sampel. So the total drift appears in x- and y-direction and is not constant.
I have some defects on the sampel which I use for the registration process. On these the approx. 15 pictures should be matched.The images themself are exact 150x150 pixel, with a frame that is not relevant the total size is 200x200.

The images should not be scaled, so rigid body transformation would be the best, but I don't know the angle of the rotation. The software should compute an approximate angel, and the landmarks from the source should fit those of the target. When I use scaled rotation or affine most of my 10 pictures which have to match the target image get grey and strange. By strange I mean that the contrast changes and the scaling procedure splits solid lines up in fragments. On automatic mode not much happens to the source images. In manual mode the matching of the landmarks looks good (examined with pictures in a stack), but the quality of the received pictures is poor.
So I think for my problem refinment of the landmarks between source and target would be great, but in manual mode that is not integrated. Batch mode is not avaiable for me, I am using Windows, think that's the reason. Also changing the quality from fast to accurate doesn't result in any real changes with the image quality as described above.

Perhaps anybody can held me with these problems, otherwise I will have to try another software. Any proposals? For the peopel who are really interested in my problem, below are two image files to "try at home" :-) 433 should be the target and 437 the source.




Thank you, Jonas

[Joel Sheffield]

Hi,

I imported the two images into ImageJ, and then converted to a stack of two
8-bit images.  I then ran StackReg, rather than TurboReg.  Initially, there
was no change, because I included the frames of the images as well.
Clearly, the superimposition of the labels took precedence over the image
detail.  When I cropped out the frame info, I got a quick realignment of
the two images such that the particles were superimposed.  You might try
making a stack of all of your 15 images and running Stack Reg on the set.

Joel


On Fri, Mar 8, 2013 at 9:35 AM, JoLi <[hidden email]>wrote:

> Hello everybody
>
> First some information about my images, which are rgb images received via
> matlab. I transform them to 8 or 16 Bit grayscale images using imagej. I
> have a slight drift and rotation of my sampel. So the total drift appears
> in
> x- and y-direction and is not constant.
> I have some defects on the sampel which I use for the registration process.
> On these the approx. 15 pictures should be matched.The images themself are
> exact 150x150 pixel, with a frame that is not relevant the total size is
> 200x200.
>
> The images should not be scaled, so rigid body transformation would be the
> best, but I don't know the angle of the rotation. The software should
> compute an approximate angel, and the landmarks from the source should fit
> those of the target. When I use scaled rotation or affine most of my 10
> pictures which have to match the target image get grey and strange. By
> strange I mean that the contrast changes and the scaling procedure splits
> solid lines up in fragments. On automatic mode not much happens to the
> source images. In manual mode the matching of the landmarks looks good
> (examined with pictures in a stack), but the quality of the received
> pictures is poor.
> So I think for my problem refinment of the landmarks between source and
> target would be great, but in manual mode that is not integrated. Batch
> mode
> is not avaiable for me, I am using Windows, think that's the reason. Also
> changing the quality from fast to accurate doesn't result in any real
> changes with the image quality as described above.
>
> Perhaps anybody can held me with these problems, otherwise I will have to
> try another software. Any proposals? For the peopel who are really
> interested in my problem, below are two image files to "try at home" :-)
> 433
> should be the target and 437 the source.
>
> <http://imagej.1557.n6.nabble.com/file/n5002064/433_pol._corr.png>
> <http://imagej.1557.n6.nabble.com/file/n5002064/437_pol._corr.png>
>
> Thank you, Jonas
>
>
>
> --
> View this message in context:
> http://imagej.1557.n6.nabble.com/Quality-of-Turboreg-output-not-sasisfying-tp5002064.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--


Joel B. Sheffield, Ph.D
Department of Biology
Temple University
Philadelphia, PA 19122
Voice: 215 204 8839
e-mail: [hidden email]
URL:  http://astro.temple.edu/~jbs

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[JoLi]

Hello Joel,

I also had already tried to work without the frame. turboreg had the same problems, that it lowered the contrast of some images. So that does not only depend on the frame. With stackreg it works better, but the first image does not get alligened with the others. Maybe I have to do this one by hand.

I'll try it again and post the result.

[JoLi]

okay... so it finally works. Except of the first picture of the series all images are aligned very well with the rigid body transformation. I have to adjust the first picture manually via turboreg and than I have the whole serie of pictures aligned.

Thank you all very much!

Jonas