Connected components

[hanslovsky]

This PR adds a new version of connected component analysis

• Special case scan line implementation for DiamondShape with unit range
• Use union find data structure to track connected components.
• No ExecutorService required (useful for running wihin parallelization frameworks like Spark)
• Use imglib2 Shape to allow for arbitrary neighborhoods.
• Restrict to integer size images because of array based union find. For larger images, we could use a sparse union find based on hash maps but this is slower and will also be limited to images that do not have more than Integer.MAX_VALUE foreground pixels. For large pixels, the task should be divided and parallelized anyway.

I suggest we wait until a new release of pom-imglib2 before merging so we can avoid overriding the managed version of imglib2 dependency.

[hanslovsky]

OK to merge @axtimwalde @tpietzsch @StephanPreibisch?

[imagejan]

This should be defined in a property, no?

<properties>
  <imglib2.version>4.6.1</imglib2.version>
</properties>

<dependency>
    <groupId>net.imglib2</groupId>
    <artifactId>imglib2</artifactId>
    <version>{imglib2.version}</version>
</dependency>

Besides this, it can probably dropped entirely when switching to the latest pom-imglib2 parent.

[hanslovsky]

Good catch @imagejan
This was necessary at the time of the PR, but we don't need this anymore, now.

[ctrueden]

If you need to override the version of a component in the future, please use a version property as @imagejan described, rather than hardcoding it in the <version> itself. Otherwise, the melting-pot script and other tooling will be unable to override it and some kinds of version skew will not be caught by the unified builds.

I filed scijava/scijava-maven-plugin#16 to hopefully aid with avoiding this scenario in the future.

[hanslovsky]

Will do (actually, already doing that in most of my projects).

[hanslovsky]

@axtimwalde has a use case in which it would be helpful if connected components analysis would produce the same id as if connected components analysis was run on the full volume for any fully contained connected component in an arbitrary subvolume. As an example, let

final RandomAccessibleInterval< B > mask = ...
final RandomAccessibleInterval< B > subVolume1 = Views.interval( mask, ... );
final RandomAccessibleInterval< B > subVolume2 = Views.interval( mask, ... );

Any connected component that is fully contained in subVolume1 should have the same label as it would have in mask. As a consequence, if that same connected component is fully contained in subVolume2, as well, it will be labeled with the same id in that sub-volume, too.

I implemented this extension in a separate branch:
https://github.com/hanslovsky/imglib2-algorithm/tree/connected-components-more-general
Example:
https://github.com/hanslovsky/sandbox-with-ij/blob/master/src/main/kotlin/org/janelia/saalfeldlab/labels/ConnectedComponentsExample.kt
If @axtimwalde thinks that this extension helps his use case, I will rebase that extension into this PR.

[hanslovsky]

@axtimwalde I rebased that branch as mentioned above.