Construction of a fine-grained retrieval model for archival text-image based on the integration of scene graph generation and attention mechanism

Mengyuan Zhang · 2026

Read Construction of a fine-grained retrieval model for archival text-image based on the integration of scene graph generation and attention mechanism by Mengyuan Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0353505.txt.

Description

Zhang, in a 2026 paper, designs a search engine for archive images that contain text, such as stamped official documents. Archives hold mountains of such pages, and finding one page by its content is hard when seals cover text, print blocks are tiny, and scans are noisy. Plain text reading misses how words and pictures relate.

The proposed system has five layers, from cleaning the image to matching a query. A scene graph step maps the elements on the page and how they relate. An attention step then fuses three kinds of features: the graph, the text, and links between image and text. On test data, the first returned result was right far more often than with a classic text-only method, including on pages with seal cover and tiny print. Each search answered in a small fraction of a second. Zhang presents the design as support for archives going online.

Catalog record

CREATORMengyuan Zhang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69884
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0353505

Browse the full library · About the Tower of Babel