DataFrames for image analysis

[lintool]

Currently, we have RDD-based analytics for image analysis here:
https://archivesunleashed.org/aut/#image-analysis

Let's DataFrame-ify it - that is, build the DataFrame infrastructure that wold support image analysis.

I'm thinking of creating two separate DataFrames:

• The first for the image links, something like (source page, image url)
• The second for the images themselves, something like (img url, type, height, width, md5, raw bytes)

[jwli229]

I will work on this.

[lintool]

@JWZ2018

Great. BTW - one end game would be to create something like this in a single e2e pipeline: http://ruebot.net/elxn42.html

Fork the repo, start working on a branch - send an initial PR when you're ready and we can discuss iteratively.

[ruebot]

Here's the background on how I make those.

[jwli229]

@lintool
Just making sure I'm on the right track and understanding the codebase properly, for the above two dataframes, I'm planning to add extractImageLinksDF and extractImageDetailsDF to the WARecordRDD in io/archivesunleashed/package.scala and then add them to DataframeLoader similar to how extractValidPagesDF and extractHyperlinksDF are implemented.
Does that sound right?

[lintool]

Yes, that's a good start!

[jwli229]

@lintool
Clarifying for the second df:

• type means png/jpg?
• what is a good way to find the height and width of the image? I did not find it in the WARCRecord or the Warc file format from https://iipc.github.io/warc-specifications/specifications/warc-format/warc-1.0/. Maybe it can be computed from the raw bytes?

[lintool]

• type means standard MIME types (for now)
• yes, we'll need to do image analysis directly - we've been using ImageMagick via the Java API: https://www.imagemagick.org/script/api.php

However, I'm not opposed to poking around for other options... I found this, for example: http://imglib2.net/

Might be a better option, as opposed to messing with JNI.

@ruebot has experience with ImageMagick - thoughts?

[ruebot]

You can get the image info with Apache Tika, which we already use in the project with language and mime type extraction. https://tika.apache.org/1.7/formats.html#Image_formats

[lintool]

With #226 this is done. Closing.