STYLE: Minor rewording the explanation of image registration in example#364
Conversation
The presence of noise is not really necessary for image registration.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
|
In my humble opinion, the previous definition is shorter, sweeter, and identifies the problem better. Distinguishing between differences due the transformation and noise usually core to the task. |
| "metadata": {}, | ||
| "source": [ | ||
| "**Image registration** finds the *spatial transformation that aligns images in the presence of noise*." | ||
| "**Image registration** is the task of finding the *spatial transformation that aligns two images with each other*." |
There was a problem hiding this comment.
In my humble opinion, the previous definition is shorter, sweeter, and identifies the problem better. Distinguishing between differences due the transformation and noise usually core to the task.
Thanks for your feedback, Matt. First of all, this is not a show-stopper to me, of course. If you don't want it to be merged, please press "Request changes", or even "Close pull request" without merging.
Anyway, my main aim is to get rid of "in the presence of noise". I don't think "the presence of noise" is essential to image registration. Image registration is often used to register baseline and follow-up, or to register two different modalities. But we can also register artificially generated images. Even if the images would be entirely free of noise, it would still be valid to call it "image registration".
The example itself does not elaborate any further on the noise that is supposed to be present. So I don't think it's essential to the example either.
We have another definition of image registration at https://github.com/SuperElastix/elastix/blob/9f87f8943ac5c90bf5fffa1b5ac1566a607ad020/README.md#L17
the task of finding a spatial transformation, mapping one image (the fixed image) to another (the moving image), by optimizing relevant image similarity metrics.
The proposed rewording would bring the text from the example more in sync with this definition from SuperElastix/elastix.
There was a problem hiding this comment.
@N-Dekker yes, I agree that the methods can be used for artificially generated images. I think it is helpful to emphasize that the quest when working natural images, which I think is the most common end use case for elastix. We are
optimizing relevant image similarity metrics
where the optimization and similarity metric has to distinguish what content is expected to correspond and what content is not expected to correspond (noise).
I do not have strong attachments myself; I am sharing my preference and how I think "noise" should be understood and its significance in this context.
3 participants
The presence of noise is not really necessary for image registration.
@mstaring @stefanklein Please check! It's just a matter of definition 😃
(Honestly this pull request also an excuse to see if the tests still pass today, at the CI. Especially because my attempt to upgrade elastix (pull request #363) has a test failure!)