• 한국멀티미디어학회논문지
• /
• 제21권2호
• /
• pp.173-180
• /
• 2018

This paper proposes a method to improve the performance of face recognition via super-resolution method using sparse representation and deep learning from low-resolution facial images. Recently, there have been many researches on ultra-high-resolution images using deep learning techniques, but studies are still under way in real-time face recognition. In this paper, we combine the sparse representation and deep learning to generate super-resolution images to improve the performance of face recognition. We have also improved the processing speed by designing in parallel structure when applying sparse representation. Finally, experimental results show that the proposed method is superior to conventional methods on various images.

• Current Optics and Photonics
• /
• 제7권1호
• /
• pp.73-82
• /
• 2023

In this study, we demonstrate ultrahigh-resolution spectral-domain optical coherence tomography with a 200-kHz line rate using a superluminescent diode with a -3-dB bandwidth of 100 nm at 849 nm. To increase the line rate, a subset of the total number of camera pixels is used. In addition, a partial-spectrum detection method is used to obtain OCT images within an imaging depth of 2.1 mm while maintaining ultrahigh axial resolution. The partially detected spectrum has a flat-topped intensity profile, and side lobes occur after fast Fourier transformation. Consequently, we propose and apply the super-Gaussian window function as a new window function, to reduce the side lobes and obtain a result that is close to that of the axial-resolution condition with no window function applied. Upon application of the super-Gaussian window function, the result is close to the ultrahigh axial resolution of 4.2 ㎛ in air, corresponding to 3.1 ㎛ in tissue (n = 1.35).

• 대한전자공학회논문지SD
• /
• 제46권5호
• /
• pp.39-48
• /
• 2009

1984년 처음 SR 알고리즘이 제안된 이후, 많은 SR 복원 알고리즘이 제안되었다 SR의 접근방법 중에서도 공간적 접근방법은 저해상도 이미지의 픽셀 값을 고해상도 이미지 격자에 매핑 함으로써 이루어진다. 이때, 저해상도 이미지들 간의 각각 다른 노이즈와 다른 PSF(Point Spread Function) 함수, 왜곡으로 인해 매핑 시 문제가 된다. 때문에 저해상도 이미지들의 노이즈 성분을 최소화하는 방법이 필요하다. 본 논문에서는 노이즈 성분을 최소화하는 방법으로 L1 norm의 방법을 사용하고 이와 동시에 이미지의 경계를 보완해주는 Huber norm을 사용하는 SR의 구조를 제안한다. 실험에서는 타 알고리즘과의 비교를 통해 제안한 알고리즘이 저해상도 이미지 상에 존재하는 노이즈를 줄이고 이미지 경계부분의 보완을 확인하였다.

• 한국융합학회논문지
• /
• 제11권12호
• /
• pp.15-22
• /
• 2020

본 논문은 채널 강조(Channel Attentin)와 공간 강조(Spatial Attention) 방법을 결합한 딥 러닝 기반의 초해상도 방법을 제안하였다. 초해상도 과정에서 질감, 특징과 같은 주변 픽셀의 변화량이 큰 고주파 성분의 복원이 중요하다. 채널 강조와 공간 강조를 결합한 특징 강조를 이용한 초해상도 방법을 제안하였다. 기존의 CNN(Convolutional Neural Network) 기반의 초해상도 방법은 깊은 네트워크의 학습이 어려우며, 고주파 성분의 강조가 부족하여 윤곽선이 흐려지거나 왜곡이 발생한다. 문제를 해결하기 위해 스킵-커넥션(Skip Connection)을 적용한 채널 강조와 공간 강조를 결합한 강조 블록과 잔차 블록(Residual Block)을 사용하였다. 방법으로 추출한 강조된 특징 맵을 부-픽셀 컨볼루션(Sub-pixel Convolution)을 통해 특징맵을 확장하여 초해상도를 진행하였다. 이를 통해 기존의 SRCNN과 비교하여 약 PSNR는 5%, SSIM은 3% 향상되었으며 VDSR과 비교를 통해 약 PSNR는 2%, SSIM은 1% 향상된 결과를 보였다.

• 융합신호처리학회논문지
• /
• 제22권4호
• /
• pp.156-163
• /
• 2021

어텐션(Attention) 메커니즘은 딥러닝 기술을 활용한 다양한 컴퓨터 비전 시스템에서 활용되고 있으며, 초해상화(Super-resolution)를 위한 딥러닝 모델에도 어텐션 메커니즘을 적용하고 있다. 하지만 어텐션 메커니즘이 적용된 대부분의 초해상화 기법들은 Real 영상의 초해상화에만 초점을 맞추어서 연구되어, 어텐션 메커니즘을 적용한 초해상화가 CG나 VR 영상 초해상화에도 유효한지는 알기 어렵다. 본 논문에서는 최근에 제안된 어텐션 메커니즘 모듈인 BAM(Balanced Attention Mechanism) 모듈을 12개의 초해상화 딥러닝 모델에 적용한 후, CG나 VR 영상에서도 성능 향상 효과를 보이는지 확인하는 실험을 진행하였다. 실험 결과, BAM 모듈은 제한적으로 CG나 VR 영상의 초해상화 성능 향상에 기여하였으며, 데이터 특징과 크기, 그리고 네트워크 종류에 따라 성능 향상도가 달라진다는 것을 확인할 수 있었다.

• 한국멀티미디어학회논문지
• /
• 제24권2호
• /
• pp.178-185
• /
• 2021

Three-dimensional high-resolution magnetic resonance imaging (MRI) provides fine-level anatomical information for disease diagnosis. However, there is a limitation in obtaining high resolution due to the long scan time for wide spatial coverage. Therefore, in order to obtain a clear high-resolution(HR) image in a wide spatial coverage, a super-resolution technology that converts a low-resolution(LR) MRI image into a high-resolution is required. In this paper, we propose a super-resolution technique through filter learning based on information on the surrounding gradient information in 3D space from 3D MRI images. In the learning step, the gradient features of each voxel are computed through eigen-decomposition from 3D patch. Based on these features, we get the learned filters that minimize the difference of intensity between pairs of LR and HR images for similar features. In test step, the gradient feature of the patch is obtained for each voxel, and the filter is applied by selecting a filter corresponding to the feature closest to it. As a result of learning 100 T1 brain MRI images of HCP which is publicly opened, we showed that the performance improved by up to about 11% compared to the traditional interpolation method.

• 한국인공지능학회지
• /
• 제12권1호
• /
• pp.1-6
• /
• 2024

Recently, various techniques are being applied through the development of medical AI, and research has been conducted on the application of super-resolution AI models. In this study, evaluate the results of the application of the super-resolution AI model to brain CT as the basic data for future research. Acquiring CT images of the brain, algorithm for brain and bone windowing setting, and the resolution was downscaled to 5 types resolution image based on the original resolution image, and then upscaled to resolution to create an LR image and used for network input with the original imaging. The SRCNN model was applied to each of these images and analyzed using PSNR, SSIM, Loss. As a result of quantitative index analysis, the results were the best at 256×256, the brain and bone window setting PSNR were the same at 33.72, 35.2, and SSIM at 0.98 respectively, and the loss was 0.0004 and 0.0003, respectively, showing relatively excellent performance in the bone window setting CT image. The possibility of future studies aimed image quality and exposure dose is confirmed, and additional studies that need to be verified are also presented, which can be used as basic data for the above studies.

• 한국진공학회:학술대회논문집
• /
• 한국진공학회 2014년도 제46회 동계 정기학술대회 초록집
• /
• pp.403.1-403.1
• /
• 2014

In conventional far-field microscopy, two objects separated closer than approximately half of an emission wavelength cannot be resolved, because of the fundamental limitation known as Abbe's diffraction limit. During the last decade, several super-resolution methods have been developed to overcome the diffraction limit in optical imaging. Among them, super-resolution optical fluctuation imaging (SOFI) developed by Dertinger et al [1], employs the statistical analysis of temporal fluorescence fluctuations induced by blinking phenomena in fluorophores. SOFI is a simple and versatile method for super-resolution imaging. However, due to the uncontrollable blinking of fluorophores, there are some limitations to using SOFI for several applications, including the limitations of available blinking fluorophores for SOFI, a requirement of using a high-speed camera, and a low signal-to-noise ratio. To solve these limitations, we present a new approach combining SOFI with speckle pattern illumination to create illumination-induced optical fluctuation instead of blinking fluctuation of fluorophore.. This technique effectively overcome the limitations of the conventional SOFI since illumination-induced optical fluctuation is possible to control unlike blinking phenomena of fluorophore. And we present the sub-diffraction resolution image using SOFI with speckle illumination.

• Journal of Electrical Engineering and Technology
• /
• 제9권1호
• /
• pp.363-371
• /
• 2014

In this paper, a new super-resolution algorithm is proposed by using successive frames for generating high-resolution frames with better quality than those generated by other conventional interpolation methods. Generally, each frame used for super-resolution must only have global translation and motions of sub-pixel unit to generate good result. However, the newly proposed MSR algorithm in this paper is exempt from such constraints. The proposed algorithm consists of three main processes; motion estimation for image registration, normalization of motion vectors, and pattern analysis of edges. The experimental results show that the proposed algorithm has better performance than other conventional algorithms.

• ETRI Journal
• /
• 제32권3호
• /
• pp.390-394
• /
• 2010

In this paper, we propose a new super-resolution technique based on interpolation of the high-frequency subband images obtained by discrete wavelet transform (DWT) and the input image. The proposed technique uses DWT to decompose an image into different subband images. Then the high-frequency subband images and the input low-resolution image have been interpolated, followed by combining all these images to generate a new super-resolved image by using inverse DWT. The proposed technique has been tested on Lena, Elaine, Pepper, and Baboon. The quantitative peak signal-to-noise ratio (PSNR) and visual results show the superiority of the proposed technique over the conventional and state-of-art image resolution enhancement techniques. For Lena's image, the PSNR is 7.93 dB higher than the bicubic interpolation.