registration tools

[Kenneth Sloan-2]

I seem to have confused myself.  I don't remember how I got into this confusion, so I'm asking
for guidance.

We do a lot of image registration.  For stacks of images captured using the same methods, we generally
use SIFT.  But, we also deal with multi-modal images, where gray-levels may be reversed, and "matching
features" don't obviously "match".  In those cases, we rely on user input to select common landmarks.  
Often, the images to be registered are different types (RGB, 8-bit, Float) and sizes.

Some time ago, I wrote a plugin that accepts 1, 2, or 3 landmarks in each image, and uses "barycentric" interpolation
to define a mapping from one image to the other.  When 1 common landmark exists, only Translation is used; when 2, I use Translation, Rotation, and Scale; when 3 are supplied, I do a more general transformation which maps the two triangles
onto each other exactly.  (Actually, I always match two triangles, but when only 1 or 2 common landmarks are provided I construct artificial landmarks to create a complete triangle in each image).

Virtually always, it turned out that 2 landmarks (giving a TRS transformation) was what we needed.  I then "discovered"
the "align with Line Roi" plugin.  This has worked just fine.

Now, however, I can't tell WHERE I found it - and I'm having trouble giving instructions to users who want to install
everything we need on their many machines.

So...is this plugin included in the standard FIJI distribution?  Or, must it be manually added?

Second - when I search for it, I find it, with documentation that says it only handles gray images.  The one I have
installed handles RGB images (and seems to change the source image to match the target).  Hence, my confusion...

While I have you - I would appreciate expert guidance as to the "best" tool for this job.  Again, the requirements are:

a) handles all image types (preferably by NOT requiring or enforcing all images to be the same type)
b) uses user input to provide 1,2,3, (or more?) matching landmarks

And, of course, where to find your preferred plugin.

Note: for image STACKS, I suppose all images will already be the same size and type.  For multi-modal images, my
preference is to register the images one-at-a-time against a single master image.  Changing the SIZE of the
source image to match the target is OK, but I would prefer to NOT be forced to match image TYPE.  But...the images
will probably eventually end up in a Stack, so it's not too onerous to use a common type (as long as RGB is supported).
In general, I prefer RGB for all color images, and tend to prefer Float for grayscale images.

My fallback will be to revive and harden my BarycentricRegistration plugin - but I'm trying to use "standard" tools where available.

--
Kenneth Sloan
[hidden email]
Vision is the art of seeing what is invisible to others.

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

[Curtis Rueden]

Hi Kenneth,

> is this plugin included in the standard FIJI distribution?

Yes, "Align Image by line ROI" is part of Fiji.

If you type "align image" into the search bar, you will find that the menu
path is Plugins > Registration > Align Image by line ROI.

You will also see that the containing JAR file is (as of this writing)
plugins/VIB_-3.0.2.jar.

Clicking the "Source" button opens the source code on GitHub:
https://github.com/fiji/VIB/blob/VIB_-3.0.2/src/main/java/Align_Image.java

Clicking the "Help" button opens the wiki page:
https://imagej.net/Align_Image_by_line_ROI

From the source, it seems that all of uint8, uint16, float32 and rgb are
supported. From the git history, we can see that support for color images
was indeed added:

https://github.com/fiji/VIB/commit/5b7b8ccffd549dc08cc4b63877e7d3bfd14698b3

It appears the documentation is simply outdated. It is a wiki—please feel
warmly encouraged to correct the error.

Regards,
Curtis

--
Curtis Rueden
LOCI software architect - https://loci.wisc.edu/software
ImageJ2 lead, Fiji maintainer - https://imagej.net/User:Rueden
Have you tried the Image.sc Forum? https://forum.image.sc/


On Fri, Dec 28, 2018 at 5:29 PM Kenneth Sloan <[hidden email]>
wrote:

> I seem to have confused myself.  I don't remember how I got into this
> confusion, so I'm asking
> for guidance.
>
> We do a lot of image registration.  For stacks of images captured using
> the same methods, we generally
> use SIFT.  But, we also deal with multi-modal images, where gray-levels
> may be reversed, and "matching
> features" don't obviously "match".  In those cases, we rely on user input
> to select common landmarks.
> Often, the images to be registered are different types (RGB, 8-bit, Float)
> and sizes.
>
> Some time ago, I wrote a plugin that accepts 1, 2, or 3 landmarks in each
> image, and uses "barycentric" interpolation
> to define a mapping from one image to the other.  When 1 common landmark
> exists, only Translation is used; when 2, I use Translation, Rotation, and
> Scale; when 3 are supplied, I do a more general transformation which maps
> the two triangles
> onto each other exactly.  (Actually, I always match two triangles, but
> when only 1 or 2 common landmarks are provided I construct artificial
> landmarks to create a complete triangle in each image).
>
> Virtually always, it turned out that 2 landmarks (giving a TRS
> transformation) was what we needed.  I then "discovered"
> the "align with Line Roi" plugin.  This has worked just fine.
>
> Now, however, I can't tell WHERE I found it - and I'm having trouble
> giving instructions to users who want to install
> everything we need on their many machines.
>
> So...is this plugin included in the standard FIJI distribution?  Or, must
> it be manually added?
>
> Second - when I search for it, I find it, with documentation that says it
> only handles gray images.  The one I have
> installed handles RGB images (and seems to change the source image to
> match the target).  Hence, my confusion...
>
> While I have you - I would appreciate expert guidance as to the "best"
> tool for this job.  Again, the requirements are:
>
> a) handles all image types (preferably by NOT requiring or enforcing all
> images to be the same type)
> b) uses user input to provide 1,2,3, (or more?) matching landmarks
>
> And, of course, where to find your preferred plugin.
>
> Note: for image STACKS, I suppose all images will already be the same size
> and type.  For multi-modal images, my
> preference is to register the images one-at-a-time against a single master
> image.  Changing the SIZE of the
> source image to match the target is OK, but I would prefer to NOT be
> forced to match image TYPE.  But...the images
> will probably eventually end up in a Stack, so it's not too onerous to use
> a common type (as long as RGB is supported).
> In general, I prefer RGB for all color images, and tend to prefer Float
> for grayscale images.
>
> My fallback will be to revive and harden my BarycentricRegistration plugin
> - but I'm trying to use "standard" tools where available.
>
> --
> Kenneth Sloan
> [hidden email]
> Vision is the art of seeing what is invisible to others.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

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

[Kenneth Sloan-2]

Thanks very much.  I thought that was the case, but my "customer" was confused.

Now...2 common landmarks are usually sufficient for my needs.  3 doesn't seem to add much, but
"n" common landmarks might be useful, as long as it is implemented well.

Is there concensus on the best existing plugin to register two images using an arbitrary
number of common landmarks?

For single-modalities, using SIFT to do it automatically seems to be my favorite,
but it doesn't work all that well in multi-modal imaging where the appearance of
"interesting points" differs wildly from modality to modality.
So, I'm set in the case where SIFT works, but interested
in recommendations on "best practice" for user-specified landmarks.

--
Kenneth Sloan
[hidden email]
Vision is the art of seeing what is invisible to others.





> On 28 Dec 2018, at 19:42 , Curtis Rueden <[hidden email]> wrote:
>
> Hi Kenneth,
>
>> is this plugin included in the standard FIJI distribution?
>
> Yes, "Align Image by line ROI" is part of Fiji.

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

[Saalfeld, Stephan]

Yes:

https://imagej.net/Landmark_Correspondences

Happy 2019,
Stephan

On Fri, 2018-12-28 at 22:39 -0600, Kenneth Sloan wrote:

> Thanks very much.  I thought that was the case, but my "customer" was
> confused.
>
> Now...2 common landmarks are usually sufficient for my needs.  3
> doesn't seem to add much, but
> "n" common landmarks might be useful, as long as it is implemented
> well.
>
> Is there concensus on the best existing plugin to register two images
> using an arbitrary
> number of common landmarks?
>
> For single-modalities, using SIFT to do it automatically seems to be
> my favorite,
> but it doesn't work all that well in multi-modal imaging where the
> appearance of
> "interesting points" differs wildly from modality to modality.
> So, I'm set in the case where SIFT works, but interested
> in recommendations on "best practice" for user-specified landmarks.
>
> --
> Kenneth Sloan
> [hidden email]
> Vision is the art of seeing what is invisible to others.
>
>
>
>
>
> > On 28 Dec 2018, at 19:42 , Curtis Rueden <
> > [hidden email]> wrote:
> >
> > Hi Kenneth,
> >
> > > is this plugin included in the standard FIJI distribution?
> >
> > Yes, "Align Image by line ROI" is part of Fiji.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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