Automated, Dynamic Decision of Image Quality Criteria
Motivation
The selection of suitable images for a given image style (also known as a “briefing”) from a large number of source images, e.g. from a photo shoot, always involves a large amount of time for the agency or repro staff involved. For this reason, methods are being developed on the basis of artificial intelligence for the automatic evaluation of image data according to application-specific image styles. Target-precise image data characteristic for a specific style (original and corrected images) are the basis for the development of a neural network that enables dynamic image quality evaluation without a formalized (human) description of the respective taste or image style. The evaluation is done in such a way that arbitrary images for a given output process are automatically checked as to whether they correspond to a certain image style and how much effort, if any, is estimated for a necessary retouching (image correction) (“traffic light system”).
Scientific Questions
Our working hypothesis is that “Image-to-Image-Translation” as well as “Image-to-Feature-Translation” are able to develop such a traffic light system. Since the decisions according to the suitability of image data are already made manually by experts, sufficient training data is available. This approach is further based on the hypothesis that the recognition of the relevant features for the gradual match with an image style is sufficient to successfully apply this style to new images – and thus to correct them automatically. This last step is particularly demanding and represents the greatest scientific challenge.
Theses
New theses are regularly advertised in the area of Automated, Dynamic Decision of Image Quality Criteria. In addition to the general overview, there are also numerous topics that have not yet been advertised, which will be gladly presented in a personal conversation.
Data-Upload
For research into new processes, we are building up an image database that contains both original, unprocessed image files and processed ones in order to optimize them for a specific image style. Such image styles can be from the fields of fashion, tourism or food, for example. In addition, we request an evaluation of these image files according to quality criteria such as (1 – poorly suited, …, 5 – perfectly suited).
Under this link you can upload your pictures and support us in our project.
Partners
- Dr. Andreas Kraushaar, Research Institute for Media Technologies e. V. (Fogra), Aschheim b. Munich
External Funding
- AiF – German Federation of Industrial Research Associations, “”Machine learning framework for dynamic decision of image quality criteria.”, Project nr. AiF 19811 N