• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.48 no.1
• /
• pp.54-63
• /
• 2011

This paper proposes a fast learning-based interpolation algorithm to up-scale an input low-resolution image into a high-resolution image. In conventional learning-based super-resolution, a certain relationship between low-resolution and high-resolution images is learned from various training images and a specific high frequency synthesis information is derived. And then, an arbitrary low resolution image can be super-resolved using the high frequency synthesis information. However, such super-resolution algorithms require heavy memory space to store huge synthesis information as well as significant computation due to two-dimensional matching process. In order to mitigate this problem, this paper presents one-dimensional patch-based learning and synthesis. So, we can noticeably reduce memory cost and computational complexity. Simulation results show that the proposed algorithm provides higher PSNR and SSIM of about 0.7dB and 0.01 on average, respectively than conventional bicubic interpolation algorithm.

• Journal of the Korean earth science society
• /
• v.32 no.6
• /
• pp.525-536
• /
• 2011

In any applications using various types of spatial data, it is very important to account for the scale differences among available data sets and to change the scale to the target one as well. In this paper, we propose to use a geostatistical downscaling approach based on vaiorgram deconvloution and block simulation to generate fine-scale categorical thematic maps from coarse-scale fraction data. First, an iterative variogram deconvolution method is applied to estimate a point-support variogram model from a block-support variogram model. Then, both a direct sequential simulation based on area-to-point kriging and the estimated point-support variogram are applied to produce alternative fine-scale fraction realizations. Finally, a maximum a posteriori decision rule is applied to generate the fine-scale categorical thematic maps. These analytical steps are illustrated through a case study of land-cover mapping only using the block fraction data of thematic classes without point data. Alternative fine-scale fraction maps by the downscaling method presented in this study reproduce the coarse-scale block fraction values. The final fine-scale land-cover realizations can reflect overall spatial patterns of the reference land-cover map, thus providing reasonable inputs for the impact assessment in change of support problems.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.71-72
• /
• 2019

본 논문에서는 계층구조 합성곱 신경망 기반의 고해상도 동영상 프레임 고속 보간 방법을 제안한다. 기존의 고해상도 동영상 프레임 보간 방법은 시간 해상도와 공간 해상도를 분리하여 보간 하기 때문에, 예측된 보간 프레임이 블러(blur) 열화를 갖는 문제를 보인다. 제안하는 방법에서는 이러한 문제를 해결하기 위하여 계층구조 합성곱 신경망 기반의 보간 방법을 이용한다. 제안하는 계층구조 합성곱 신경망은 우선 저해상도의 광학 흐름 추정지도를 생성하고 이를 고해상도로 복원하여 프레임 보간을 수행한다. 이때, 저해상도 광학 흐름 지도를 추정할 때 사용된 특징 정보들을 활용하여 고품질의 고해상도 광학 흐름 지도를 추정한다. 실험을 통하여 제안하는 방법이 고해상도 프레임을 고속으로 보간하며, 동시에 블러 열화에 대한 성능 향상을 가짐을 보였다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38C no.10
• /
• pp.875-880
• /
• 2013

When we generate 3D video that provides immersive and realistic feeling to users, depth information of the scene is essential. Since the resolution of the depth map captured by a depth sensor is lower than of the color image, we need to upsample the low-resolution depth map for high-resolution 3D video generation. In this paper, we propose a depth upsampling method using depth-discontinuity information. Using the high-resolution color image and the low-resolution depth map, we detect depth-discontinuity regions. Then, we define an energy function for the depth map upsampling and optimize it using the belief propagation method. Experimental results show that the proposed method outperforms other depth upsampling methods in terms of the bad pixel rate.

• Journal of the Korea Society of Computer and Information
• /
• v.16 no.6
• /
• pp.71-77
• /
• 2011

In this paper, we propose a learning based super resolution algorithm using local block for image enhancement of vehicle license plate. Local block is defined as the minimum measure of block size containing the associative information in the image. Proposed method essentially generates appropriate local block sets suitable for various imaging conditions. In particular, local block training set is first constructed as ordered pair between high resolution local block and low resolution local block. We then generate low resolution local block training set of various size and blur conditions for matching to all possible blur condition of vehicle license plates. Finally, we perform association and merging of information to reconstruct into enhanced form of image from training local block sets. Representative experiments demonstrate the effectiveness of the proposed algorithm.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2014.06a
• /
• pp.107-108
• /
• 2014

부화소 기반 영상 축소기법은 각각의 부화소를 조절함으로써 픽셀 기반 영상 축소기법보다 해상도를 향상시킬 수 있는 방법이다. 그러나 부화소에 의한 해상도의 증가는 종종 색상정보의 왜곡을 발생시킨다. 부화소 랜더링의 주요과제는 선명도를 유지함과 동시에 색조왜곡현상을 억제하는 것이다. 선행연구들은 부화소랜더링을 위해 1 차원 혹은 2 차원 필터를 최적화 하였지만, 지역적인 특성을 고려하지 않았기 때문에 출력영상의 화질이 저하되는 현상이 발생한다. 본 논문은 위와 같은 문제를 해결하기 위해 내용적응형 2D FIR 필터를 제작방법을 제안한다. 제안필터는 충분한 수의 저해상도 패치와 고해상도 패치 쌍을 이용하여 임의의 고해상도 패치로부터 고화질의 저해상도 패치를 만들기 위한 최적의 내용적응형 2D FIR 필터를 학습한다. 학습된 필터에 의한 실험결과 제안하는 필터가 종례기법들 보다 색조왜곡현상이 현저히 줄어들고, 출력영상의 선명도를 유지함을 보여준다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.5-8
• /
• 2019

단일 영상 초해상도 (Single Image Super-Resolution - SISR)기법은 카메라로 획득된 저해상도 영상에 필터 기반의 연산을 적용하여 좋은 화질의 고해상도 영상을 복원하는 과정이다. 최근에 심층 합성곱 신경망 학습의 발전에 따라 단일 영상 초해상도에 적용되는 심층 학습 기법들은 좋은 성과를 보여 주고 있다. 본 논문은 단일 영상 초해상도 성능을 개선하기 위해 웨이블릿 예측 네트워크를 효율적으로 적용하는 방법에 대해 연구하였으며, 저해상도 입력 영상의 특징을 잘 추출해내기 위해 네트워크 내부에 RDB를 적용하여 기존 방식보다 효율적으로 고해상도 영상 복원하는 기법을 제안한다. 모의실험을 통해 제안하는 방법이 기존 방법보다 화질은 약 PSNR 0.18dB만큼 우수하며 속도는 1.17배 빠른 것을 확인하였다.

• Korean Journal of Remote Sensing
• /
• v.29 no.1
• /
• pp.69-79
• /
• 2013

This paper presents a two-stage geostatistical integration approach that aims at downscaling of coarse scale remote sensing data. First, downscaling of the coarse scale sedoncary data is implemented using area-to-point kriging, and this result will be used as trend components on the next integration stage. Then simple kriging with local varying means that integrates sparse precise observation data with the downscaled data is applied to generate thematic information at a finer scale. The presented approach can not only account for the statistical relationships between precise observation and secondary data acquired at the different scales, but also to calibrate the errors in the secondary data through the integration with precise observation data. An experiment for precipitation mapping with weather station data and TRMM (Tropical Rainfall Measuring Mission) data acquired at a coarse scale is carried out to illustrate the applicability of the presented approach. From the experiment, the geostatistical downscaling approach applied in this paper could generate detailed thematic information at various finer target scales that reproduced the original TRMM precipitation values when upscaled. And the integration of the downscaled secondary information with precise observation data showed better prediction capability than that of a conventional univariate kriging algorithm. Thus, it is expected that the presented approach would be effectively used for downscaling of coarse scale data with various data acquired at different scales.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.48 no.9
• /
• pp.22-30
• /
• 2011

Recently, the rising demand for intelligent video surveillance system leads to high-performance face recognition systems. The solution for low-resolution images acquired by a long-distance camera is required to overcome the distance limits of the existing face recognition systems. For that reason, this paper proposes a hardware design of an image resolution enhancement algorithm for real-time intelligent video surveillance systems. The algorithm is synthesizing a high-resolution face image from an input low-resolution image, with the help of a large collection of other high-resolution face images, called training set. When we checked the performance of the algorithm at 32bit RISC micro-processor, the entire operation took about 25 sec, which is inappropriate for real-time target applications. Based on the result, we implemented the hardware module and verified it using Xilinx Virtex-4 and ARM9-based embedded processor(S3C2440A). The designed hardware can complete the whole operation within 33 msec, so it can deal with 30 frames per second. We expect that the proposed hardware could be one of the solutions not only for real-time processing at the embedded environment, but also for an easy integration with existing face recognition system.

• Proceedings of the Korean Information Science Society Conference
• /
• 2004.10b
• /
• pp.802-804
• /
• 2004

번호판 인식시스템의 개발에 있어서 번호판 영역의 추출단계는 시스템의 성능에 큰 영향을 미치는 단계이며 문자인식단계 이상으로 중요하다. 본 논문에서는 웹 카메라를 이용하여 얻어진 저해상도 영상으로부터 번호판 고유의 색상과 텍스쳐를 이용하여 번호판영역을 추출하고, 허프변환을 이용한 기울어진 영상의 회전을 통해 번호판 문자 영역화 및 인식에 용이한 차량번호판 영상을 추출하는 기법을 제안한다.