이방성 확산을 이용한 DCT 기반 압축 영상의 블록효과 제거

최종성(Jongseong Choi),최은철(Euncheol Choi),강문기(Moon Gi Kang) 한국방송·미디어공학회 2002 한국방송공학회 학술발표대회 논문집 Vol.2002 No.-

한국종교학회 40년의 길

최종성(Choi, Jongseong) 한국종교학회 2011 宗敎硏究 Vol.64 No.-

적응적 가중치를 이용한 노이즈에 강인한 초점값 연산자

최종성(Jongseong Choi),강희(Hee Kang),강문기(Moon Gi Kang) 대한전자공학회 2010 電子工學會論文誌-SP (Signal processing) Vol.47 No.4

초점 검출 시스템은 영상 획득 과정에서 영상의 화질을 결정하는 중요한 요소이다. 초점 검출은 크게 영상의 고주파 성분을 평가하여 수치화하는 초점값 연산 부분과 이 초점값을 이용하여 렌즈를 이동시켜 초점을 일치시키는 부분으로 이루어진다. 초점값을 연산하는데 있어 저조도 잡음이 첨가된 환경에서는 잡음에 의해 그 성능이 크게 저하되게 된다. 본 논문에서는 공간 적응적인 가중치를 이용하여 저조도 잡음이 첨가된 환경에서 효율적으로 초점값을 연산할 수 있도록 하는 방법을 제안하였다. 제안된 방법은 영상의 각 픽셀에서 영상의 국부 특성과 잡음의 특성을 적응적 가중치를 연산하고, 이를 이용해 저조도 잡음에 강인한 초점값 연산자를 제안한다. 제안된 적응적 가중치는 기존의 필터 기반 초점값 연산자에도 적용이 가능한 특성을 갖는다. 잡음이 없는 상태와 가우시안 잡음이 있는 환경 하에서 제안된 연산자의 성능을 검증하였다. The focusing system is an important factor to determine the imaging quality of a digital imaging system. The focusing system consist of measuring the focusing index with high frequency energy of an image and controlling the movement of the focusing lens based on the computed focusing index. The computation of the focusing index is a key aspect in implementing the focusing system and the noise of the image cause the error in the sharpness evaluation of the image. To reduce this error, the noise under the low illumination condition is considered. A noise insensitive focusing index using adaptive weights is proposed in this paper. This measure determines the sharpness of an image using the spatially adaptive weights based on the local statistics of the image and noise. Experimental results under the condition without and with the noise verify the performance of the proposed method.

東學의 신학과 인간학

최종성(Choi, Jongseong) 한국종교학회 2006 宗敎硏究 Vol.44 No.-

ARIO: 지진으로부터의 학습하는 대용량 시각데이터 자동분류기

최종성(Jongseong Choi),Shirley J. Dyke 대한기계학회 2022 대한기계학회 춘추학술대회 Vol.2022 No.3

복수 노출을 이용한 공간 해상도와 다이내믹 레인지 향상 알고리즘

최종성(Jongseong Choi),한영석(Young-seok Han),강문기(Moon Gi Kang) 대한전자공학회 2008 電子工學會論文誌-SP (Signal processing) Vol.45 No.6

이미지 센서의 물리적 한계 가운데 공간 해상도의 제약과 다이내믹 레인지의 제약을 극복하기 위한 방법 가운데 신호처리기법에 기반하여 여러 장의 저해상도 영상으로부터 고해상도 영상을 복원하는 것과, 다이내믹 레인지가 좁은 여러 장의 영상으로부터 넓은 다이내믹 레인지를 갖는 영상을 복원하는 방법이 있다. 하지만, 일반적으로 실제 영상을 획득하는 과정에서 공간 해상도와 다이내믹 레인지의 제약을 동시에 받게 되므로, 이 두 제약을 동시에 극복하는 연구가 필요하다. 본 논문에서는 영상 장치의 응답 함수의 추정과 함께 공간 해상도와 다이내믹 레이지를 동시에 향상시킬 수 있는 알고리즘을 제안한다. 이를 위해 영상의 공간 해상도 제한과, 다이내믹 레인지의 제약을 포함하는 영상 획득 과정을 모델링하고, 이 영상 획득 모델을 기반으로 하여 영상 입력 장치의 응답 함수를 추정하고, 영상의 공간 해상도와 다이내믹 레인지를 동시에 향상시킬 수 있는 알고리즘을 제안한다. 실험 결과를 통해 제안된 알고리즘이 기존의 고해상도 복원 알고리즘과 와이드 다이내믹 레인지 영상 복원을 연속적으로 처리한 결과보다 시각적, 수치적으로 더 좋은 결과를 보여줌을 확인할 수 있다. The approaches to overcome the limited spatial resolution and the limited dynamic range of image sensors have been studied independently. A high resolution image is reconstructed from multiple low resolution observations and a wide dynamic range image is reconstructed from differently exposed multiple low dynamic range images based on signal processing approach. In practical situations, it is reasonable to address them in a unified context because the recorded image suffers from limitations of both spatial resolution and dynamic range. In this paper, the image acquisition process including limited spatial resolution and limited dynamic range is modelled. With the image acquisition model, the response function of the imaging system is estimated and the single image of which spatial resolution and dynamic range are simultaneously enhanced is obtained. Experimental results indicate that the proposed algorithm outperforms the conventional approaches that perform the high resolution and wide dynamic range reconstruction sequentially with respect to both objective and subjective criteria.

NeRF기반 디지털 트윈 환경의 심리스 3D 장면 제어

임한솔(Hansol Lim),최종성(Jongseong Choi) 대한기계학회 2023 대한기계학회 춘추학술대회 Vol.2023 No.11

This paper introduces a method for achieving seamless 3D scene through NeRF generated Digital Twin environment and its implementation in Unity. NeRF (Neural Radiance Fields), a specialized neural network technique, is used to create a photorealistic Digital Twin from sparse point cloud data. This data is derived from Structure-from-Motion environmental images. The NeRF Digital Twin is subsequently exported to Unity to be rendered in a virtual environment in real-time, which is then displayed on a screen. The proposed methodology utilizes camera geometry and world coordinate transformations to create a control system that dynamically aligns the virtual environment to the user’s perspective for a seamless 3D view without a use of a VR gear. The screen displays the synthesized 3D view aligned to the user’s perspective. The proposed approach offers a practical solution for creating a view-through effect on a screen, without the use of glass or mirrors and future-proof structures for mixed reality applications.

실시간 로그데이터 분석 기반의 전기차 경로 최적화 기법

조나단 보약(Jonathan Boyack),최종성(Jongseong Choi) 대한기계학회 2022 대한기계학회 춘추학술대회 Vol.2022 No.11

An issue presented when the question arises of why more people are not using electric vehicles is that the support structure for EVs is not as widespread as for gas-powered cars. For this reason, path optimization for long-distance travel including quick charging and minimizing travel time has been a crucial task for development. To improve upon current methods, we use real-time log data to glean information about the vehicle’s physical properties, and the driver’s abilities and timing to give an optimal path. Here, the log data is a record of the voltage and current with the associated vehicle speed in realtime. Using such data, we have developed a method that takes these things into account and produces an optimized path between two points while connecting them with quick charge locations where the vehicle would be low on battery. If the vehicle has been loaded more or less than usual or if the drag of the vehicle is affected which can cause the initially planned path to need correction, our method does this automatically.

실시간 로그데이터 분석 기반의 전기차 경로 최적화 기법

조나단 보약(Jonathan Boyack),최종성(Jongseong Choi) 대한기계학회 2022 대한기계학회 춘추학술대회 Vol.2022 No.11

An issue presented when the question arises of why more people are not using electric vehicles is that the support structure for EVs is not as widespread as for gas-powered cars. For this reason, path optimization for long-distance travel including quick charging and minimizing travel time has been a crucial task for development. To improve upon current methods, we use real-time log data to glean information about the vehicle’s physical properties, and the driver’s abilities and timing to give an optimal path. Here, the log data is a record of the voltage and current with the associated vehicle speed in realtime. Using such data, we have developed a method that takes these things into account and produces an optimized path between two points while connecting them with quick charge locations where the vehicle would be low on battery. If the vehicle has been loaded more or less than usual or if the drag of the vehicle is affected which can cause the initially planned path to need correction, our method does this automatically.

Falling Hazard Assessment Application: Automated Labelling Large Data Sets for Transfer Learning Models

Jonathan Boyack(조나단 보약),Jongseong Choi(최종성) 대한기계학회 2023 대한기계학회 춘추학술대회 Vol.2023 No.11

Natural disasters frequently cause serious damage to business, buildings, and homes. This damage creates areas around the structure that are dangerous to people or vehicles moving nearby. To clear these areas for workers to approach, an inspector must complete an evaluation of the exterior of the building to warn of hazardous areas. These inspections can be lengthy depending on building size and design which can inhibit the daily life of those in the affected areas. By combining novel techniques in machine learning with new image sensors and platforms, such as unmanned aerial vehicles (UAVs), the time these inspections take can be greatly reduced. In this paper we explained the method developed which uses a convolutional neural network (CNN) which has been trained on a small number of labeled images using transfer learning to determine whether the images are damaged or safe. The images are also used to generate an orthophoto of the exterior of the building using a technique called structure from motion (SfM). Over each location of damage on a portion of the orthophoto, a region of interest (ROI) is generated. Depending on how large a portion of the orthophoto is covered those ROIs are transferred to the smaller images and used for training data for the CNN. This decreases the amount of labelling required for the training data set.