Quickly select the best photos with Deep Learning

The smartphone has fundamentally changed photography. The permanent availability of the camera built into the phone allows users to capture almost every moment of their lives. Day after day, gigantic amounts of images and data are created in this way. The photo service provider CEWE relies on neural networks to support its customers in viewing and evaluating these photos in the future.

Looking at photos on the smartphone

Far more pictures are taken today by smartphone than by digital camera. At the same time, users' shooting habits have changed. Special experiences and unforgettable moments are still being photographed. However, mobile phone owners are taking photos more often for purely informative or communicative purposes - the photo becomes a visual reminder or a short message for friends. The flood of images on the devices is correspondingly large; it continues to grow through sharing via app or social media. Due to the sheer volume, the photos cannot be curated, organized or designed afterwards. Even important photos disappear into the depths of ever-increasing storage media - and are rarely printed out.

AI recognizes relevant and well made images

The photo service provider CEWE wants to support hobby photographers in coping with the flood of data - with the help of artificial intelligence. The Oldenburg company had already launched software for designing photo books on the market in 2006. It was based on algorithms that were not only able to identify the technically best pictures on a device, but also - with the help of intelligent clustering - the most relevant ones. However, automatically finding the most relevant images and determining and presenting meaningful events can no longer be achieved with classic methods based on algorithms and heuristics. CEWE therefore relies on Deep Learning. Above all, this requires relevant data in large quantities, as is produced sufficiently in mobile photography.

The AI applications developed by CEWE focus on neural networks, which are trained, for example, to determine the best image of a series. Basic rules of photographic design theory are applied, such as the rule of thirds, which is based on the proportion theory of the golden section. Even pictures with relevant people and places can be detected automatically. Using face recognition, photos with people are automatically combined. The object recognition used is based on an extensive thesaurus and groups matching images, e.g. according to keywords such as "beach" or "mountains". Location recognition rounds off the image organization through geographic clustering. Images of desired persons, in a favorite setting and with suitable temporal parameters can be searched and filtered. For the customer this means in perspective: a better selection of photos in less time.

Positive feedback from the industry

The new technology is still under development and not commercially viable. Together with its development partner, OFFIS - Institut für Informatik, also located in Oldenburg, CEWE has already defined and tested the first use cases. At the photokina photo trade fair in Cologne, these generated a positive response. Since a completely AI-controlled solution for image selection and processing still needs time, CEWE is currently pursuing a two-track approach: individual partial solutions can be integrated into the applications available on the market in advance.

The use of Artificial Intelligence requires responsibility in an area as personal as photography. For this reason, CEWE has adopted a customer charter. It states that in all technical solutions - current and future ones - the benefit of the human being and the protection of privacy are always in the foreground. Whether the customer wants to use the technical possibilities is up to him: Those who want to edit their images without the help of AI can deselect the corresponding option.

CEWE Stiftung & Co. KGaA

AI Users
Large Enterprise
Website

AI Development partner
OFFIS - Institut für Informatik

Application facts


Technology field
Image Recognition and Understanding
Data Management and Analysis
Natural Language Processing
Application industry
Other Services
Field of application
  • Intelligent Assistance Systems
Value-added activity
Service/Customer Service

Video