• Journal of Digital Contents Society
• /
• v.15 no.6
• /
• pp.737-743
• /
• 2014

For single image super resolution (SR), interpolation based and example based algorithms are extensively used. The interpolation algorithms have the strength of theoretical simplicity. However, those algorithms are tending to produce high resolution images with jagged edges, because they are not able to use more priori information. Example based algorithms have been studied in the past few years. For example based SR, the nearest neighbor based algorithms are extensively considered. Among them, neighbor embedding (NE) has been inspired by manifold learning method, particularly locally linear embedding. However, the sizes of local training sets are always too small. So, NE algorithm is weak in the performance of the visuality and quantitative measure by the poor generalization of nearest neighbor estimation. An improved NE algorithm with Support Vector Regression (SVR) was proposed to solve this problem. Given a low resolution image, the pixel values in its high resolution version are estimated by the improved NE. Comparing with bicubic and NE, the improvements of 1.25 dB and 2.33 dB are achieved in PSNR. Experimental results show that proposed method is quantitatively and visually more effective than prior works using bicubic interpolation and NE.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.437-450
• /
• 2022

Generative adversarial networks(GANs) have demonstrated remarkable success in image synthesis. However, since GANs show instability in the training stage on large datasets, it is difficult to apply to various application fields. A single image GAN is a field that generates various images by learning the internal distribution of a single image. In this paper, we investigate five Single Image GAN: SinGAN, ConSinGAN, InGAN, DeepSIM, and One-Shot GAN. We compare the performance of each model and analyze the pros and cons of a single image GAN.

• Journal of Broadcast Engineering
• /
• v.26 no.4
• /
• pp.429-440
• /
• 2021

With the recent development of deep convolutional neural network learning, deep learning techniques applied to single image super-resolution are showing good results. One of the existing deep learning-based super-resolution techniques is RDN(Residual Dense Network), in which the initial feature information is transmitted to the last layer using residual dense blocks, and subsequent layers are restored using input information of previous layers. However, if all hierarchical features are connected and learned and a large number of residual dense blocks are stacked, despite good performance, a large number of parameters and huge computational load are needed, so it takes a lot of time to learn a network and a slow processing speed, and it is not applicable to a mobile system. In this paper, we use the residual dense structure, which is a continuous memory structure that reuses previous information, and the residual dense channel attention block using the channel attention method that determines the importance according to the feature map of the image. We propose a method that can increase the depth to obtain a large receptive field and maintain a concise model at the same time. As a result of the experiment, the proposed network obtained PSNR as low as 0.205dB on average at 4× magnification compared to RDN, but about 1.8 times faster processing speed, about 10 times less number of parameters and about 1.74 times less computation.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2016.11a
• /
• pp.117-118
• /
• 2016

최근 UHDTV(ultra high definition television)가 가정에 보급이 많이 되고 있는 추세지만, UHD급 콘텐츠가 매우 부족한 실정이다. 따라서 저해상도 FHD(full high definition) 영상을 고해상도 영상으로 변환시켜 재활용할 수 있는 초해상화(super-resolution, SR) 기술의 필요성이 커졌다. 그 중, 다층의 레이어로 구성된 다층 선형 매핑(multi-layer linear mappings, MLLM)을 기반으로 하는 제안된 초해상화 기법은 상대적으로 낮은 복잡도로 좋은 품질의 고해상도 영상을 복원할 수 있었다. 최근에는 강화 예측법을 추가하여 복원된 고해상도 영상의 품질을 더 향상시키는 기법이 등장하였는데, 이를 바탕으로 본 논문에서는 제안했었던 MLLM 기법을 위한 강화 예측법 기법을 새롭게 제안한다. 제안하는 초해상화 기법은 기존 MLLM 기법과 딥러닝 기반 초해상화 기법보다 높은 품질의 고해상도 영상을 생성하는 것을 확인하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.226-227
• /
• 2020

근래의 초해상화 (super-resolution, SR) 연구는 네트워크를 깊고, 넓게 만들어 성능을 높이는데 주를 이뤘다. 그러나 동시에 높은 연산량과 메모리 소비량이 증가하는 문제가 발생하기 때문에 이를 실제로 하드웨어로 구현하기에는 어려운 문제가 존재한다. 그렇기에 우리는 네트워크 최적화를 통해 성능 감소를 최소화하면서 파라미터 수를 줄이는 네트워크 SqueezeSR을 설계하였다. 또한 지식 증류(Knowledge Distillation, KD)를 이용해 추가적인 파라미터 수 증가 없이 성능을 높일 수 있는 학습 방법을 제안한다. 또한 KD 시 teacher network의 성능이 보다 student network에 잘 전달되도록 feature map 간의 비교를 통해 학습 효율을 높일 수 있었다. 결과적으로 우리는 KD 기법을 통해 추가적인 파라미터 수 증가 없이 성능을 높여 다른 SR네트워크 대비 더 빠르고 성능 감소를 최소화한 네트워크를 제안한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.325-326
• /
• 2020

고해상도의 홀로그램을 얻기 위한 다양한 연구가 지속되고 있다. 본 논문은 고해상도의 위상 홀로그램을 획득하기 위하여 딥러닝 기반의 학습과 복원 결과를 가지고 분석을 진행한다. 사용된 위상 홀로그램은 보편적인 이미지와 값의 범위가 동일하다. SISR(Single Image Super Resolution)에서 좋은 결과를 보인 네트워크를 사용하여 위상 홀로그램에 대한 학습을 진행하였다. 네트워크로 획득한 홀로그램과 원본 홀로그램의 복원 결과를 비교하여, 차이점과 개선해야할 것들에 대해서 심도 있게 분석한다.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.2 s.308
• /
• pp.12-18
• /
• 2006

This paper presents a method to extract a good high-resolution image from a low-resolution video in an automatic manner. Since a high-resolution image reconstruction method utilizing several low-resolution input images works better than a conventional interpolation method utilizing single low-resolution input image only if the input images are well registered onto a common high-resolution grid, low-resolution input images should be carefully chosen so that the registration errors can be carefully considered. In this paper, the statistics obtained from the motion-compensated low-resolution images are utilized to evaluate the feasibility of the input image candidates. Maximum motion-compensation error is estimated from the high-resolution image observation model. U the motion-compensation error of the input image candidate is greater than the estimated maximum motion-compensation error, the input image candidate is discarded. The number of good input image candidates and the statistics of the motion-compensation errors are used to choose final input images. The final input images chosen from the input image selection block are given to the following high-resolution image reconstruction block. It is expected that the proposed method is utilized to extract a good high-resolution image efficiently from a low-resolution video without any user intervention.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.255-256
• /
• 2019

본 논문에서는 초고해상도 기술의 복잡도를 줄이기 위하여 픽셀단위 생략 방법을 제안한다. 제안하는 방법은 픽셀 단위로 수평, 수직 방향의 밝기에 대한 2 차 미분치에 기반하여 생략을 결정한다. 제안하는 방법의 성능 평가를 위하여 가장 간단한 초고해상도 알고리즘인 SRCNN 과 제안하는 방법의 PSNR 비교한다. 그 결과 제안하는 방법이 평균적으로 약 47%의 픽셀이 생략이 되면서 SRCNN 대비 0.2dB PSNR 감소를 보인다.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.378-390
• /
• 2022

With the recent development of deep composite multiplication neural network learning, deep learning techniques applied to single-image super-resolution have shown good results, and the strong expression ability of deep networks has enabled complex nonlinear mapping between low-resolution and high-resolution images. However, there are limitations in applying it to real-time or low-power devices with increasing parameters and computational amounts due to excessive use of composite multiplication neural networks. This paper uses blocks that extract hierarchical characteristics little by little using information distillation and suggests the Recursive Distillation Super Resolution Network (RDSRN), a lightweight network that improves performance by making more accurate high frequency components through high frequency residual purification blocks. It was confirmed that the proposed network restores images of similar quality compared to RDN, restores images 3.5 times faster with about 32 times fewer parameters and about 10 times less computation, and produces 0.16 dB better performance with about 2.2 times less parameters and 1.8 times faster processing time than the existing lightweight network CARN.

• Journal of Digital Contents Society
• /
• v.18 no.8
• /
• pp.1517-1522
• /
• 2017

Among single image SR techniques, the TV based SR approach seems most successful in terms of edge preservation and no artifacts. But, this approach achieves insufficient SR for texture component. In this paper, we proposed a new TV-G decomposition based SR method to solve this problem. We proposed the SVR based up-sampling to get better edge preservation in the structure component. The NNE used the relaxed constraint to improve the NE. We used the NNE based learning method to improve the resolution of the texture component. Through experimental results, we quantitatively and qualitatively confirm the improved results of the proposed SR method when comparing with conventional interpolation method, ScSR, TV and NNE.