Artificial intelligence and the automatic classification of historical photographs

The amount of historical photographs is a challenge for archives with regard to analyzing, documenting and structuring archival material and make it retrievable and accessible. Established human centric concepts will not be able to handle the instream of un- or semi-structured material or handle analogue documentation. We suggest the use of artificial intelligence to pre-classify and organize historic photographs in collections and archives. With regard to uncertainty the concept of artificial intelligence based classification is discussed based on two questions: (1) What is the uncertainty in current archival documentation, assuming that the quality and completeness of a current classification depends on the partial and restricted knowledge of a person? (2) Can algorithmic concepts help to quantify the level of uncertainty? The aspect of documentation is discussed based on a comparison of algorithms and classical processes. The contribution of artificial intelligence to quantify the uncertainty is tested in a case study. For this an artificial neural network / convolutional neural network was trained and tested on a collection of historical photographs.