Content AI solution company FourByFour announced on the 15th that a new AI image quality enhancement method paper written by its in-house AI research lab has been accepted as a paper for the international conference ‘ICCE 2025’ hosted by the Institute of Electrical and Electronics Engineers (IEEE).
The paper written by FourByFour AI Lab ‘Pixell Lab’ covers the topic of ‘Spatiotemporal Recurrent Alignment Method for Video Quality Enhancement.’ By improving the previously used linear process into a recurrent process that repeats by segments for video quality enhancement, it is possible to maintain quality while reducing time and cost.
The existing method for improving video quality involved arranging images within the video linearly and capturing areas to be enhanced through changes between preceding and succeeding frames, i.e., the ‘optical flow estimation’ method. This conventional approach requires multiple computational steps, making it complex and time-consuming. It also suffers from considerable errors due to incorrect estimations, and the detected defects require a separate, high-cost motion compensation process.
The method proposed by FourByFour arranges video frames in a recurrent structure rather than linearly and blocks them, enabling quick detection of changed and improvable areas without complex computations. This reduction in computational processing shortens the video enhancement time, reduces errors and noise, and improves video consistency, thereby maximizing video quality.
Kim Hyuncheol, head of AI research at FourByFour and author of the paper, said, "Because we approached this from the perspective of company members who actually sell and service AI solutions for image quality enhancement, rather than researchers developing AI models, we were able to find this improvement method." He added, "Having confirmed that such a simple process improvement by changing the frame alignment method can achieve this level of efficiency, it will greatly help actual product performance and service improvements."
The conference ‘ICCE (International Conference on Consumer Electronics)’ where the paper was accepted is an international academic event held alongside CES, one of the world’s top three consumer electronics exhibitions, introducing AI and image processing technologies. As an annual flagship conference hosted by the IEEE Consumer Technology Society (IEEE CTSoc), which boasts the highest academic credibility in this field worldwide, it is an international event where top experts from industry, government, and academia gather.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
