Award-winning processing technique cuts out noise from real-world images
Modern electronic devices are capable of producing high-quality images, but the process of acquiring and processing images is complex. Sometimes, disruptions in signals that produce images lead to “noise,” which causes the images to be blurry or of poor quality. To correct these “noisy” images, one type of approach involves the addition of synthetic noise to a noise-free image. But, this method is flawed, as these artificial noisy images are very different from real-word images. To focus on real-world images, another type of approach involves the use of an image acquisition algorithm, in which a ground-truth image (reflecting how reality looks) is converted to an artificially produced noisy image. The noisy, blurry image can then be reverted to the original, sharp image using the same algorithm. This technique is effective, but currently, it lacks a large enough dataset to denoise images efficiently.
In a new study published in the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, a team of researchers at Dongguk University in South Korea, led by Dr Seung-Won Jung, built a state-of-art image denoising method, which is much more efficient than previous methods. Talking about the importance of this study, Dr Jung says, “Real-world image denoising is still very challenging because it is not possible to obtain ideal pairs of ground-truth and real-world noisy images. Through our study, we wanted to resolve the issues faced by existing denoising methods.”
For their study, the scientists used images from the database of the NTIRE 2019 Real Image Denoising Challenge. By generating a new computational architecture, which they named as “group residual dense network,” they managed to augment the existing dataset of real-world noisy and ground-truth images, which helped them achieve the highest possible level of peak signal-to-noise ratio and structural similarity. What’s more, this new image-correcting method was even ranked first place at the NTIRE 2019 Real Image Denoising Challenge for its novelty and efficiency. Describing this method, Dr Jung says, “While previous research often used artificially produced noisy images, we built a database of real-world noisy images instead.”
Removing noise is important for devices such as smartphones and digital cameras to produce high-quality images. By finding a more efficient denoising method, this study has possibly made significant advancements in the field of image processing. Dr Jung concludes by talking about the applications of the study, “We believe that our real-world noise modeling method can be extended to various applications, such as removing blur, aliasing, and haze in images produced by devices like mobile phones and cameras, showing its versatility.”
Title of original paper: GRDN: Grouped Residual Dense Network for Real Image Denoising and GAN-based Real-world Noise Modeling
Author: Dr Seung-Won Jung
Affiliation: Department of Multimedia Engineering, Dongguk University, Seoul
About Dongguk University
Dongguk University, founded in 1906, is located in Seoul, South Korea. It comprises 13 colleges that cover a variety of disciplines and has local campuses in Gyeongju, Goyang, and Los Angeles. The university has 1,300 professors who conduct independent research and 18,000 students undertaking studies in a variety of disciplines. Interaction between disciplines is one of the strengths on which Dongguk prides itself; the university encourages researchers to work across disciplines in Information Technology, Bio Technology, CT, and Buddhism.
About the author
Dr Seung-Won Jung completed his BS and PhD degrees in electrical engineering from Korea University, Seoul, Korea, in 2005 and 2011, respectively, where he was a Research Professor with the Research Institute of Information and Communication Technology in Korea University from 2011 to 2012. He was a Research Scientist at the Samsung Advanced Institute of Technology, Yongin, Korea, from 2012 to 2014. He is currently an Assistant Professor at the Department of Multimedia Engineering, Dongguk University, Seoul. He has authored over 50 peer-reviewed articles in international journals. His current research interests include augmented reality, virtual reality, and computer vision algorithms.