Extracted Illustrations of the Berlin State Library's Digitized Collections (part 1 of 4)

The dataset consists of various illustrations extracted from 26,233 historical books and other media offered in the Berlin State Library's Digitized Collections. The media objects are older than 1920.

Version 1.0 contains of 594,890 extracted illustrations in total.

The extraction of illustrations is driven by the coordinates given by the ABBYY FineReader OCR engine (in ALTO XML) . The extracted illustrations have not been resized but compressed and saved in JPEG format.

Pre-trained models in order to separate color scales, hand-written signatures, library stamps or the like from interesting content are available under: https://github.com/elektrobohemian/imi-unicorns.

The extracts for each media object are stored in separated sub-folders and tar files named after the PPN (a unique ID used in the library) to facilitate further processing. Additional metadata can be obtained with help of the PPN as described here: https://github.com/elektrobohemian/StabiHacks/blob/master/ppn-howto.md .

The dataset is published as a set of ZIP files, each fitting on a Blu Ray disc. After decompression, the contents will consume ca. 166 GB.

Change Log for Version

1. original dataset
2. added color histograms (RGB, separated by channel) in JSON and Python pickle format as extracted by the Pillow package (see https://github.com/elektrobohemian/StabiHacks/tree/master/image-tools)

This is part 1 of 4. The following datasets contain the other ZIP files (8 files in total):

• https://doi.org/10.5281/zenodo.2598145
• https://doi.org/10.5281/zenodo.2598261
• https://doi.org/10.5281/zenodo.2598270