Owners of digital cameras, smartphones and computers are actively using the ability to process images with the help of special algorithms. Such filters for photos are used in popular Photoshop, Instagram and Prisma applications. Using neural networks for this purpose allows bringing the process of image filtration to a new quality level. As an example, we can name the emerging possibility of changing the style of a photo in accordance with the style of another image and with high precision.
At present, changing the style is a painstaking and lengthy process of involving experienced artists and photographers, who perform the necessary transformations manually. Automation of such processes in Prisma or Facebook applications failed to achieve the desired quality of the final image.
Specialists from Cornell University (New York, USA) together with the developers from the Adobe company managed to create a software package that allows changing the style of a digital image in an automatic mode without involving the user. The program has not been given a name yet, but the studies are conducted within the framework of the research called “Deep Photo Style Transfer”.
The operating principle of the software package is based on the already available algorithm for changing the style but involves a more active use of neural network capabilities. The program is set a task of transferring the image style from the basic photo to the photo the consumer needs. As a result of studying the style of the basic image by the neural network, an adaptation of the resulting image takes place with the maximum preservation of its appearance and structure. The developed project is offered with open source codes on the Git Hub website and will soon be included in Adobe graphics processors.