Digital photography and especially the spread of smartphones have strongly influenced the way we take pictures. For example, private photos on a smartphone are mixed up with practical shots such as shopping lists, handwritten notes, addresses or even the number of the parking space in a large parking lot. Another typical challenge is the question of which of the many pictures of a vacation are so beautiful and aesthetic that they should definitely be printed in a photo album. OFFIS follows different approaches to prepare personal picture collections in a way that the captured memories can be relived.
For this purpose, the researchers train Convolutional Neural Networks (CNNs) with the goal of automatically evaluating and labeling the photos in personal photo collections. This is an AI method with multi-layered, "deep" structures. These ratings make it easier to select beautiful images for photo services like albums or greeting cards. The CNN method used for this purpose learns from annotated data sets how people rate pictures, for example as practical or as personally valuable and aesthetic. Information such as the number of people, the overall mood, the environment in which the picture was taken and the aesthetics of the pictures can thus be captured with a certain degree of probability and used for intelligent selection and suggestions.
For many years, the cooperation between OFFIS, as representative of the research world, and the Oldenburg company CEWE, as representative of the photo services, has proven to be very fruitful. Due to the close cooperation, the research questions are closely oriented to problems and topics from the practice. With this partnership, OFFIS supports the rapid transfer of application-oriented research into commercial use in intelligent photo services.
This article was originally published in the magazine "Technologie-Information" of Leibniz Universität Hannover: https://www.uni-hannover.de/de/forschung/transfer/technologie-informationen