Image Collation: Matching Illustrations in Manuscripts

Illustrations are an essential transmission instrument. For historian, the first step in studying their evolution a corpus of similar manuscripts is to identify which ones correspond each other. This image collation task daunting for separated by many lost copies, spreading over centuries, might have been completely re-organized and greatly modified adapt novel knowledge or belief include hundreds illustrations. Our contributions this paper threefold. First, we introduce illustration large annotated public dataset evaluate solutions, including 6 2 different texts with more than 000 illustrations 1 200 correspondences. Second, analyze state art similarity measures show that they succeed simple cases but struggle when undergone very significant changes discriminated only fine details. Finally, clear evidence performance boosts can be expected exploiting cycle-consistent code data available on http://imagine.enpc.fr/~shenx/ImageCollation.