• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2016.06a
• /
• pp.9-11
• /
• 2016

최근 UHDTV(ultra high definition television) 등의 고해상도 디스플레이가 시장에 등장하면서, 기존의 저해상도 FHD(full high definition) 영상을 고해상도 영상으로 변환할 수 있는 초해상화(super-resolution, SR) 기법들이 각광을 받고 있다. 그 중, 선형 매핑(linear mapping)을 사용하여 저해상도 패치(patch)로부터 고해상도 패치를 복원하는 초해상화 기법은 상대적으로 낮은 복잡도로 좋은 품질의 고해상도 영상을 생성한다. 그러나 이러한 기법은 단순한 선형 매핑을 기반으로 하기 때문에 복잡한 비선형적(nonlinear) 저해상도-고해상도 관계를 예측하기 힘든 단점이 있다. 최근 각광받는 딥러닝(deep learning) 기술은 다층(multi-layer) 네트워크를 쌓아 입력과 출력 간의 복잡한 비선형 관계를 훈련시켜 좋은 성능을 보이는데, 이를 바탕으로 본 논문에서는 다중의 레이어로 구성된 다층 선형 매핑(multi-layer linear mappings, MLLM)을 기반으로 하는 초해상화 기법을 새롭게 제안한다. 제안하는 다층 선형 매핑은 기존 선형 매핑보다 비선형적 관계를 더 잘 예측하여 높은 품질의 고해상도 영상을 생성할 수 있게 한다. 제안된 초해상화 기법은 딥러닝 기반 초해상화 기법과 필적하는 품질의 고해상도 영상을 생성하면서도 더 낮은 복잡도를 지니는 것을 확인하였다.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.453-454
• /
• 2003

Most research materials (data), which are used for the study of digital mapping and digital elevation model (DEM) in the field of Remote Sensing and Aerial Photogrammetry are aerial photographs and satellite images. Additionally, they are also used for National land mapping, National land management, environment management, military purposes, resource exploration and Earth surface analysis etc. Although aerial photographs have high resolution, the data, which they contain, are not used for environment exploration that requires continuous observation because of problems caused by its coastline, as well as single - spectral and long-term periodic image. In addition to this, they are difficult to interpret precisely because Satellite Images are influenced by atmospheric phenomena at the time of photographing, and have by far much lower resolution than existing aerial photographs, while they have a great practical usability because they are mulitispectral images. The PKNU 2 is an aerial photographing system that is made to compensate with the weak points of existing aerial photograph and satellite images. It is able to take pictures of very high resolution using a color digital camera with 6 million pixels and a color infrared camera, and can take perpendicular photographs because PKNU 2 system has equipment that makes the cameras stay level. Moreover, it is very cheap to take pictures by using super light aircraft as a platform. It has much higher resolution than exiting aerial photographs and satellite images because it flies at a low altitude about 800m. The PKNU 2 can obtain multispectral images of visible to near infrared band so that it is good to manage environment and to make a classified diagram of vegetation.

• Applied Microscopy
• /
• v.46 no.4
• /
• pp.184-187
• /
• 2016

Neuronal connectivity determines brain function. Therefore, understanding the full map of brain connectivity with functional annotations is one of the most desirable but challenging tasks in science. Current methods to achieve this goal are limited by the resolution of imaging tools and the field of view. Macroscale imaging tools (e.g., magnetic resonance imaging, diffusion tensor images, and positron emission tomography) are suitable for large-volume analysis, and the resolution of these methodologies is being improved by developing hardware and software systems. Microscale tools (e.g., serial electron microscopy and array tomography), on the other hand, are evolving to efficiently stack small volumes to expand the dimension of analysis. The advent of mesoscale tools (e.g., tissue clearing and single plane ilumination microscopy super-resolution imaging) has greatly contributed to filling in the gaps between macroscale and microscale data. To achieve anatomical maps with gene expression and neural connection tags as multimodal information hubs, much work on information analysis and processing is yet required. Once images are obtained, digitized, and cumulated, these large amounts of information should be analyzed with information processing tools. With this in mind, post-imaging processing with the aid of many advanced information processing tools (e.g., artificial intelligence-based image processing) is set to explode in the near future, and with that, anatomic problems will be transformed into informatics problems.

• The Transactions of the Korean Institute of Electrical Engineers C
• /
• v.53 no.1
• /
• pp.30-35
• /
• 2004

We have carried out a basic experiment in order to develope a super high-resolution optical microscope which transcend the limitation of diffraction and the wavelength of lightwave. The image of this scope is composed by measuring the evanescent wave which is localized on the surface of the testing materials. A detecting probe was fabricated with a single mode optical fiber to be sharpened by the chemical etching, and drived by PZT. The standing wave of $0.33\mu\textrm{m}$ wavelength evanescent wave which was generated from the $0.78\mu\textrm{m}$-wavelength semiconductor laser was detected by the $0.5\mu\textrm{m}$-thickness optical fiber probe.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.42-43
• /
• 2022

본 논문은 모바일 단말에서 실시간으로 동작하는 딥러닝 기반 경량 초고해상화 기술에 관한 내용이다. 대용량 3차원 메쉬 모델의 비실시간 압축은 실시간 스트리밍 응용 시나리오에서 제약점으로 작용하고 있고, 본 논문에서는 두 입력 텐서의 차원을 일치시켜야 하는 element-wise 덧셈 대신 concatenation을 활용해 연산량을 개선하고, float-to-int8 양자화 과정에서 발생하는 오차를 줄이기 위해 weight clipping 및 regularization 기법을 활용해 초고해상화 화질 성능을 개선하였다. 제안하는 알고리즘은 기존 모바일 초고해상화 기술을 화질 측면에서 0.12dB, 처리 속도 측면에서 13.6ms 개선하였고, Mobile AI & AIM 2022 실시간 이미지 초고해상화 대회에서 1등을 수상하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.537-539
• /
• 2020

단일 영상 기반 초고해상도 복원은 컴퓨터 비전 및 영상처리 분야의 중요한 기초 및 응용 분야 중 하나이며, 딥러닝에 대한 연구가 발전됨에 따라 이를 이용한 다양한 연구들이 활발히 진행되고 있다. 기존 딥러닝 기반 연구들은 복원 성능을 높이기 위해서 다양한 구조의 네트워크를 설계하거나 네트워크를 학습하는 알고리즘들을 중점으로 연구되어 왔다. 최근 들어 네트워크 구조나 설계 이외에 네트워크를 통과하는 정보의 집합체인 특징 맵에 관한 연구들이 진행되고 있다. Attention은 특징 맵에서 채널 간의 관계를 이용하여 특정 채널을 강조하거나 또는 공간 정보를 강조하는 방식으로 특징 맵의 정보를 잘 활용하도록 하여 전체적인 네트워크의 성능을 향상시킨다. 본 논문은 단일 영상 기반 초고해상도 복원 네트워크를 기반으로 다양한 Attention방법들을 적용하고 성능을 비교 및 분석한다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.2
• /
• pp.871-888
• /
• 2020

With the development of display engineering and technology, the screen and sound quality of information devices such as TVs are improving. The screen used LEDs via LCD and PDP and a large flat panel in the early CRT to create super-high resolution. The sound is improved by directly vibrating a thin and simple panel, such as an OLED. In our previous study, the exciter speaker was attached to the rear of the OLED panel to be used as the diaphragm of the speaker, and the sound quality was as good as that of the TV using the existing dynamic speaker. This method supplied the viewer with the direct sound coming from the panel, delivering clear sound, and the sound and image came from the same location, thus giving the viewer high immersion and maximizing the effect of information transfer. OLED exciter speakers, however, have a special directivity, which tends to slightly attenuate the tone at the very center of the screen. This study improves the sound quality by improving the structure of the exciter speaker and the radiated sound of the flat panel display. A 2-in-1 Exciter is made into a single core to improve the speaker's radiation pattern.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.5
• /
• pp.185-191
• /
• 2022

In this study, we propose a novel super-resolution neural network based on U-Net, residual neural network, and sub-pixel convolution. To prevent the loss of detailed information due to the max pooling of U-Net, we propose down-sampling and connection using sub-pixel convolution. This uses all pixels in the filter, unlike the max pooling that creates a new feature map with only the max value in the filter. As a 2×2 size filter passes, it creates a feature map consisting only of pixels in the upper left, upper right, lower left, and lower right. This makes it half the size and quadruple the number of feature maps. And we propose two methods to reduce the computation. The first uses sub-pixel convolution, which has no computation, and has better performance, instead of up-convolution. The second uses a layer that adds two feature maps instead of the connection layer of the U-Net. Experiments with a banchmark dataset show better PSNR values on all scale and benchmark datasets except for set5 data on scale 2, and well represent local area patterns.