19th Century Newspaper Analytics

19th Century Newspaper Analytics


Hello: my name is Paul Fyfe. And I am Qian
Ge. We’re from North Carolina State University  and are very proud to accept the runners-up
award in the Research Category for the 2016 British 
Library Labs Awards. Our project is called Illustrated Image Analytics. We’re exploring
how  computer vision and digital image processing
can be used on nineteenth-century illustrated  newspapers. We’re using materials from the
British Library’s collection of digitized historical 
newspapers, including The Graphic, The Illustrated Police News, and the Penny Illustrated 
Paper. All of these newspapers used a new technique called woodblock engraving which
made  possible the mass printing of images before
photography. They helped to create a new visual  culture in the nineteenth century, thousands
and thousands of images circulating in print. Our 
question is, with a similarly enormous scale of digital image files, how might we use 
contemporary computer vision technologies to sort and understand historical illustrations? 
We’ve tried several different techniques: image segmentation, face detection, machine 
learning frameworks, and GIST descriptors. And we’ve learned that images made from
engraved  lines can be really hard for computers to
see, even though human eyes have no trouble with 
them. So instead of asking computers to understand what these historical images represent, we’
ve been asking computers to take simple image measurements, including pixel ratio and entropy 
levels, and to show us patterns in the data. Using this method, we end up seeing some interesting 
things, including clusters of maps, illustrations at night, portrait collections, crowd scenes,
and  the gradual appearance of halftone images
toward the end of the nineteenth century. Our work is 
still in the early stages, and you can check out our project website for more background
and  updates. Thanks to the British Library for
their support and encouragement of digital  scholarship!

Author:

Leave a Reply

Your email address will not be published. Required fields are marked *