Images shrink upon registration using BigStitcher

Hi all,
I am using the BigStitcher plugin for time-registering my data set, which is an explant from embryos, where nuclei have been imaged.

I first detected interest points in these images based on nuclei and registered first using Fast Descriptor-based (translation invariant) algorithm followed by another round of registration with the Assign Closest Point with ICP (with no invariance) algorithm.

The registration itself works great. However, after registration the entire image shrunk dramatically. Any ideas why and how to solve this problem? I am a bit clueless here.

You can download both the non-registered xml and the registered xml files here ( and you will see the difference.

Thanks in advance for the help.

1 Like

I solved the problem. When I do not fix views during registration process, for some reason the entire image shrinks. I do not understand why.

But fixing the first view and mapping back registrations to first view works perfectly.

If some one can provide an explanation, it will be great.

1 Like

Welcome to the forum @sundar

Can’t answer with any confidence because I havn’t looked at anything yet, but here I go anyway.

This can happen with affines sometimes if the goal is to make sets of points closer to each other. The algorithm can “cheat” by making everything small - so regularization is usually needed to prevent the cheating.

Will try to have a look at some point.