IIIF Curation Platform: User-Driven Image Sharing with Machine Learning-Based Image Annotation

We present the “IIIF Curation Platform” (ICP) [1] that we officially released in November 2018. The ICP was built upon the idea of user-driven image sharing called curation. Unlike provider-driven image sharing, which is typical among GLAM organizations for publishing their properties to the public, user-driven image sharing is for scholars and citizens who want to build their own virtual collections and add metadata of their own. There are no straightforward methods to realize this functionality within the current IIIF specifications because this usage is considered out of their scope.

To support the workflow of user-driven image sharing, the ICP offers seven open source components in two categories. The first category includes client-side tools for browsing and managing IIIF curations, namely 1) IIIF Curation Viewer (and its embedded version), 2) IIIF Curation Manager, 3) IIIF Curation Editor, 4) IIIF Curation Finder, and 5) IIIF Curation Player. The second category includes server-side tools for storing and serving IIIF curations, namely 1) JSONkeeper, and 2) Canvas Indexer. Those tools are designed to follow the pattern of modular design and microservices, including Docker installation, and are built with interoperability in mind, including synchronization between JSONkeeper and Canvas Indexer implemented using the preliminary version of the IIIF Change Discovery API.

We applied the ICP to several applications, such as the Collection of Facial Expressions [2], Bukan Complete Collection [3], North China Railway Archive [4], and a museum exhibition at The Museum of Art, Kochi Prefecture, each implemented by a different combination of ICP components. Open source license, together with instruction for easier installation and operation, helps increase the adoption of the ICP, such as the University of Tokyo [5].

Finally, we show how we use the ICP with machine learning for image annotation. The current usage of machine learning is two-fold. First, automatic image tagging was used for the Collection of Facial Expressions. On Canvas Indexer, we developed an API for a machine learning service (image tagging) so that Canvas Indexer requests the service to annotate fragments of images by machine generated tags and return them for tag-based search at Canvas Indexer. Second, automatic image colorization was used for the North China Railway Archive. On IIIF Curation Viewer, we added a choice between an original gray-scale photograph and a colorized photograph, using the Choice of Alternative Resources (oa:Choice) defined in IIIF Presentation API 2.1. This approach allows us to create a natural representation of a photograph with multiple expressions, namely one manifest with two resources, the original photograph and the colorized photograph, on the same canvas so that we can switch between them on the same coordinate system.

Future work includes an authentication and authorization mechanism to manage IIIF sources which are closed or semi-closed.

文献情報

Asanobu KITAMOTO, Jun HOMMA, Tarek SAIER, "IIIF Curation Platform: User-Driven Image Sharing with Machine Learning-Based Image Annotation", 2019 IIIF Conference, 2019年6月 (in English)

関連資料・関連ウェブサイト

サイト内関連ページ

| リンク 1 |