Ruprecht-Karls-Universität Heidelberg
HCI->Computer Vision Group->Research->Title

Objects in Mirror are Closer than they Appear
Objectrecognition and Automatic Comparision of the Sachsenspiegel Codices

Program for assistant Professors, Ministry of Science, Research and Art (MWK), Baden-Württemberg
Björn Ommer, Peter Bell & Antonio Monroy

Cultural heritage consists not only of innovations but also of their reproductions and variations. Therefore it is crucial to evaluate the quality of these reproductions of art as well as their stylistic and semantic changes. Especially the manuscript culture of the Middle Ages flourished through manual reproductions. A prominent example is the frequently reproduced codex of Eike von Repgow's Sachsenspiegel which was composed in the years between 1220-1235 in eastern Saxony. It constitutes an outstanding piece of medieval cultural history. Eike von Repgow's text is one of the oldest prose works written in the and the earliest German vernacular law book and thus one of the most important monuments in the history of German law.
Only four illustrated versions of the text have been preserved, these Codices picturati from the 14th century are named after their present location in Heidelberg (H), Dresden (D), Wolfenbüttel (W), and Oldenburg (0). The focus of the project is to introduce a computer-based methodology for measuring and analyzing the variability between compositions of complete scenes and individual objects of these different manuscripts.

Fig 1. The illustrations show the mapping of two illustrations from Dresden (a) and Wolfenbüttel (b)

Fig 2. In this case, the algorithm was able to match the illustration of a coin minter with the original illustration (Dresden), a very similar image (Wolfenbüttel) and also an
earlier and less similar sketch from the Heidelberg manuscript via the facsimile.

Fig 3. This indoor scene can also be found in the three reproductions from Dresden, Wolfenbüttel and Heidelberg. The latter shows up as the seventh search result. However,
one notices that other similarly composed indoor scenes also appear as search results.

Fig 4. Here, the algorithm is able to find the same scene in different reproductions (see results 1, 2 and 10). The rest of the results contain other similarly composed illustrations.

Fig 5. In order to detect specific objects more accurately, more search boxes may be drawn over the relevant sections of the image. The user can improve search results by
choosing the most significant parts of the searched object such as the legs and head of a horse.

[ People | Publications | Research | Projects | Teaching | Student Projects | HCI Home | CompVis Home ]     

Last update: 21.05.2015, 13:20
zum Seitenanfang