Digging into Image Data Digging into image data

Digging into Image Data Digging into image data to answer authorship-related questions Professor Peter Ainsworth Dr Michael Meredith 9/18/2021 © The University of Sheffield / Department of French

Digging into Image Data International Project Team University of Sheffield Peter Ainsworth Michael Meredith University of Illinois at Urbana-Champaign Peter Bajcsy Rob Kooper Robert Markley Jennifer Guliano Karen Fresco Kevin Franklin Anne D. Hedeman Tenzing Shaw Michael Simeone Michigan State University/MATRIX Dean Rehberger Wayne Dyksen Matt Geimer Justine Richardson Steve Cohen Wayne Dyksen 9/18/2021 © The University of Sheffield / Department of French

Digging into Image Data Project Summary Explore image data authorship across a large collection of data Investigate the accuracy and computational scalability of adaptive image analyses applied to diverse collections of image data while driven by the same overarching question of authorship Design image analysis algorithms that will extract salient image features, group together images based on similarity of these, classify groups according to a priori knowledge, optimise algorithmic steps and parameters Apply the algorithms jointly developed to our very large image collections Report accuracy and computational requirements over all the collections 9/18/2021 © The University of Sheffield / Department of French

Digging into Image Datasets Nine complete 15 th-c Froissart manuscripts digitised to similar standards and quality. The virtual manuscripts total over 6, 100 images mainly at 500 DPI 9/18/2021 © The University of Sheffield / Department of French Images (c) Besançon, Bibliothèque d'Etude et de Conservation Scriptura Ltd, University of Sheffield

Digging into Image Datasets 17 th- and 18 th-c map collections: the University of Illinois Library holds a 1664 Blaeu Atlas and over twenty of the Atlases published by Herman Moll in the early 18 th century, as well as digital scans of the maps for this project. 9/18/2021 © The University of Sheffield / Department of French Images (c) University of Illinois

Digging into Image Datasets 19 th- and 20 th-c quilt images: the Quilt Index (a partnership of Michigan State University and the Alliance for American Quilts) contains images and detailed information on nearly 25, 000 quilts, which will grow to 50, 000 by the end of the grant period 9/18/2021 © The University of Sheffield / Department of French Images (c) Alliance for American Quilts (www. quiltindex. org) and Michigan State University Museum

Digging into Image Data Research Methodology Break down the problem of how best to discover salient characteristics into three lowlevel semantic components characterizing image content: (1) Image representations (2) Feature descriptors (3) Machine learning methods and similarity metrics for assignment of authorship Image (c) Peter Bajcsy, University of Illinois (from grant proposal) 9/18/2021 © The University of Sheffield / Department of French

Digging into Image Data Project Outputs The final output will consist primarily of : (a) Data about the salient characteristics of an artist with respect to another artist and with respect to a group of artists (b) Software for extracting salient characteristics The data could be viewed as evidence supporting authorship assignment based on: 1) Image-derived primitives including image representation, image features and machine learning models for assigning authorship 2) Human-defined semantic descriptors of unique characteristics that map onto a combination of multiple image-derived primitives 9/18/2021 © The University of Sheffield / Department of French

Digging into Image Data Digging into image data to answer authorship-related questions Professor Peter Ainsworth: p. f. ainsworth@sheffield. ac. uk Dr Michael Meredith: M. Meredith@sheffield. ac. uk Project URL: http: //www. shef. ac. uk/hri/projects/projectpages/did_images 9/18/2021 © The University of Sheffield / Department of French