Earticle

본 연구에서는 인터벤션 시술 시 사용하는 DSA에서 발생하는 높은 선량으로 시술자와 환자의 의료 피폭이 발생됨을 인지하고 3D printing DSA 팬텀을 이용하여 Cu filter와 post-processing algorithm의 적용을 통해 화질 저하를 개선한 저선량 프로토콜이 임상에 적용 가능한지를 알아봤다. 본원에서 3D Printing 기술을 이용하여 DSA 팬텀을 자체 제작하였으며, (1) 기존 프로토콜, (2) Cu filter 적용 프로토콜, (3) Cu filter 및 post-processing algorithm 적용 프로토콜로부터 획득한 영상의 선량, SNR&CNR, 정성적인 평가를 실시했다. Cu filter 0.3mm parameter 적용 전후로 61%의 선량 감소를 가져왔고, post-processing algorithm을 함께 적용했을 때 기존 Standard 프로토콜 대비 SNR은 93%, CNR은 98% 정량적인 수치를 나타냈다. 그리고 정성평가 결과에서는 93%의 인원이 임상에서 적합한 수준의 영상이라고 답변했다. Cu filter, post-processing algorithm을 통한 의료 피폭 선량 감소와 영상 화질 개선을 확인하고 기존 프로토콜을 대체할 저선량 프로토콜의 효용성을 입증했다

In this study, we recognized the high radiation dose during digital subtraction angiography (DSA) used in intervention procedures and used 3D-printed DSA phantom to determine whether low-dose protocol with post-processing algorithm is applicable to clinical procedures without deterioration of image quality. DSA Phantom was prepared by using 3D printing technology at the Asan medical center, and three imaging acquisition protocols were tested; (1) default protocol, (2) low-dose protocol with Cu filtration, (3) low-dose protocol with Cu filtration and post-processing algorithm according to radiation dose, SNR & CNR, and qualitative evaluation. It resulted in a 61% dose reduction before and after an adaption of 0.3mm Cu filtration, with 93% in SNR and 98% in CNR compared to the existing Standard protocol. In addition, 93% of the respondents said that the quality assessment was appropriate for clinical procedures. We identified that reduced radiation exposure dose and improved image quality through Cu filter, post-processing algorithm and demonstrated the utility of low-dose protocols could replace existing standard protocols.

본 연구의 목적은, 3D Hoffman 뇌 팬텀을 이용하여 얻은 영상을 median 필터, Wiener 필터와 같이 기존의 널리 사용되는 필터외의 median modified Wiener filter (MMWF) 알고리즘을 적용하여 영상의 질을 평가하였다. 영상 평가를 위해 111MBq 의 99mTc을 3D Hoffman 뇌 팬텀에 주입 후 영상을 획득 하고, 관심영역을 설정해 contrast to noise ratio (CNR) 의 정량적 분석 방법을 사용하였다. 결과적으로 획득한 영상과 비교했을 때 MMWF 알고리즘 적용 영상의 CNR 값은 1.07 배 상승 함을 확인 하였다. 본 연구를 통해 핵의학 영상에서의 MMWF 알고리즘의 유용성 을 확인 할 수 있었다

In this study, we evaluated image performance by applying the conventional filters (median and Wiener filters) and proposed median modified Wiener filter (MMWF) algorithm which are able to improve image quality using the 3D Hoffman brain phantom with 111 MBq in 99mTc. The single photon emission computed tomopraphy (SPECT) was used to acquire the gamma image during 60 s per frame (total frames : 120). To analyze the images, contrast to noise ratio (CNR) was used. According to results, the CNR values 3.01, 3.14, 3.16, and 3.23 for images with acquisition, median filter, wiener filter, and MMWF algorithm, respectively. In conclusion, we confirmed that image with MMWF algorithm is the most efficient to improve the image quality in the nuclear medicine images.

본 연구의 목적은 전산화단층촬영 (computed tomography, CT)영상으로부터 각각의 검색창 크기가 설정된 3차원 비지역적 평균 (3D non-local means, 3DNLM) 알고리즘을 적용하여 이에 대한 영상 특성 및 시간 분해능의 경향성을 분석하는 것이다. 이를 위해 Gaussian 및 poisson 노이즈가 부가된 3D Shepp-Logan phantom 영상을 모델링한 후 평활화 상수 및 패치크기가 각각 0.01 및 3×3×3으로 설정된 3DNLM 알고리즘의 검색창 크기를 3×3×3부터 2×2×2 간격으로 15×15×15 크기까지 변경하여 각각 적용하였다. 이에 대한 영상 특성의 정량적 평가를 위해 contrast to noise ratio (CNR) 및 peak signal to noise ratio (PSNR)을 측정하였으며, 시간 분해능 분석을 위해 3DNLM 알고리즘의 연산 시간을 측정하였다. 결과적으로, CNR은 검색창이 증가함에 따라 개선됨을 보였으며, PSNR 은 5×5×5 크기의 검색창에서 가장 개선된 결과를 보였다. 시간분해능의 경우 검색창의 크기가 증가함에 따라 지수함수 적으로 증가함이 확인되었다. 결론적으로, CT 영상으로부터 3DNLM 알고리즘을 효율적으로 적용하기 위해서 영상 특성 및 시간분해능이 모두 고려된 검색창의 크기를 설정하는 것이 중요함을 확인하였다.

The purpose of this study was to analyze the tendency of image characteristics and time resolution by applying a 3D non-local means (3DNLM) algorithm, which each search window size was set, in computed tomography (CT) images. For this purpose, the 3D Shepp-Logan phantom images were modeled by adding Gaussian and poisson noise. Then, search window size of 3DNLM algorithm with smoothing factor and patch size set to 0.01 and 3×3×3, respectively, changed from 3×3×3 to 15×15 ×15 at interval 2×2×2 and applied, respectively. To quantiative evaluate of image characteristics, contrast to noise ratio (CNR) and peak signal to noise ratio (PSNR) were measured, and the operation time of the 3DNLM was measured for time resolution analysis. As a result, the CNR showed improvement as the search window increased, and PSNR showed the most improved in the search window size of 5×5×5. In addition, time resolution was confirmed that it increased exponentially as the search window size increased. In conclusion, we confirmed that setting the search window size considering both image characteristics and time resolution to efficiently apply the 3DNLM algorithm in CT images.

문화재 복원은 자료와 전문가의 기술 및 고 기록물 자료에 의존해 복구한다. 하지만 새로운 문화재 자료 가 출토되거나, 잘못된 정보로 복원에 오류가 발생하면 다시 복원하는 데 오랜 시간이 걸린다. 본 연구에서는 인 공지능을 이용해 문화재를 복원하는 방법을 연구했다. 문화재를 복원하는 방법으로 기본에 2D 이미지 기반의 복 원 벗어나 문화재 특성상 3D형태의 복원이 적합하다는 판단으로 연구를 진행했다. 문화재 중 석탑을 기준으로 3D 데이터로 변환한 후 학습을 진행하고 복원정도를 확인했다. 제안된 문화재 복원은 데이터의 형태에 따라 2D형태 와 다르게 복잡도, 정확도, 복원 시간이 달라질 수 있다. 본 연구에서는 문화재 데이터를 3D 데이터 표현방법(3D Modeling)을 선정했고 RaSGAN 알고리즘을 수정해 진행했다. 최근에는 대부분의 복원 연구가 2D 형태의 GAN 알고리즘을 이용한 2D데이터(이미지) 복원 관련 연구되고 있지만 3D형태의 알고리즘은 없는 상태이다. 본 연구에서는 문화재 복원을 위해 3D GAN인 개선된 복셀(Voxel) 방식의 RaSGAN 통하여 학습을 진행하고 복원 가능성 확인했다.

The restoration of cultural properties depends on the data and the skills of experts and IT skills. However, there are times when new cultural assets are unearthed and restoration takes too long. And there is a possibility that the result may be different from what was expected. Therefore, we want to quickly restore cultural properties using artificial intelligence. There are two ways to restore cultural assets: a 2D deep learning method and a 3D deep learning method. 3D deep learning has a variety of ways to represent 3D data. Cultural properties restoration may vary in complexity, accuracy, and restoration time depending on input data. Therefore, the selection of the 3D data expression method and the selection of the deep learning algorithm are important. Recently, 2D data (image) restoration using GAN algorithm is being studied, and 3D restoration research is also being conducted through GAN algorithm. In this paper, 3D GAN was selected for the restoration of cultural properties. And it showed excellent performance in 2D GAN. The 3D GAN algorithm restored cultural properties by selecting the RASGAN algorithm using voxels and semantics.

오디오 산업은 프리미엄 산업으로써 나날이 발전 하고 있다. 입체 음향 시스템에 관한 연구는 많이 진행 되고 있다. 하지만 Audio database, algorithm, evaluation, metadata scheme 이 모두 각각 이루어지고 있다. 하나의 시스템에서 만들어진 audio 알고리즘을 평가 하고, 저장 할 수 있다면 입체 음향 오디오 연구 발전에 도움이 될 것이다. 그래서 이 논문 에서는 실감형 3D 오디오의 알고리 즘을 시스템 적으로 평가 할 수 있는 Database Architecture 제안 하고, 이 Database system 구현 을 위하여 XML metadata scheme를 정의 하였다. 본 논문에서는 새로운 오디오 평가 DB를 제시하 고, 이를 체계적으로 구현하기 위한 설계를 제시하고자 한다.

In this paper an architecture for 3D audio core algorithm evaluation database system. Due to increase of 3D audio system through multimedia device, an evaluation system is required for evaluating the 3D core algorithms for developing 3D audio system. Conventional evaluation systems have some problems. Researchers have to learn usage of evaluation system, in addition it is inefficient to use and search audio sources because audio sources are not indexed in general. To solve these problems, we design the architecture of 3D audio core algorithm evaluation database system enabling to automatically evaluate core algorithms using database management system. Also we define XML metadata scheme for information of saved audio source in database. This approach allows improving efficiency of search audio source and use of audio database.

지금은 인터넷 통신의 대중화와 통신 기술의 발달로 멀티미디어 콘텐츠가 많이 개발되었다. 특히 비디오 게임에 대한 시장이 커지고 있다. 사용자의 몰입감을 향상시키기 위해 사용자와의 상호작용 기능을 지원하는 멀티미디어 콘텐츠가 지속적으로 증대되고 있다. 본 논문에서는 실세계에서 사용자의 움직임을 3D 위치 정보로 변환하여 얻을 수 있는 움직임 센싱 장치인 Microsoft의 Kinect[1]와 SURF 알고리즘[2]을 이용하여 3D 가상공간에서 3차원 오브젝트를 선택하여 변환할 수 있는 3D 사용자 인터페이스 시스템을 제안한다. 제안한 시스템을 사용을 통해 몰입감과 편리성이 증대되었음을 알았다.

Recently, a lot of multimedia content have been developed due to both enhancement of communication technology and popularization of internet communication. In order to improve user's immersion, multimedia content supporting user interaction are continuously increasing. In this paper, we propose 3D user-interface systems which can pick objects in 3D virtual space and then can transform them using SURF algorithm[2] and Kinect of Microsoft[1], which is a moving sensing device obtaining 3D positions information for the movements of users in real space. Experimental results show that the immersion as well as the convenience are increased.

스테레오 매칭 과정에 있어서 매칭 비용을 구하는 것은 매우 중요한 과정이다. 이러한 스테레오 매칭 과정의 성능을 살펴보기 위하여 본 논문에서는 기존에 제안된 매칭 비용 함수들에 대한 기본 개념들을 소개하고 각각의 성능 및 장점 을 분석하고자 한다. 가장 간단한 매칭 비용 함수는 매칭 되는 영상의 일관된 밝기를 이용하여 좌, 우 영상 간 서로 대 응하는 대응점을 추정하는 과정으로, 본 논문에서 다루는 매칭 비용함수는 화소 기반과 윈도우 기반의 매칭 비용 방법 으로 크게 두 가지로 나눌 수 있다. 화소 기반의 방법으로는 절대 밝기차(the absolute intensity differences: AD) 와 sa mpling-intensitive absolute differences of Birchfield and Tomasi (BT) 방법이 있고, 윈도우 기반의 방법으로는 차이 절 대 값의 합(sum of the absolute differences: SAD), 차이 제곱 값의 합(sum of squred differences: SSD), 표준화 상호 상관성(normalized cross-correlation: NCC), 제로 평균 표준화 상호 상관성(zero-mean normalized cross-correlation: Z NCC), census transform, the absolute differences census transform (AD-Census) 이 있다. 본 논문에서는 앞서 언급 한 기존에 제안된 매칭 비용 함수들을 정확도와 시간 복잡도를 측정했다. 정확도 측면에서 AD-Census 방법이 평균적 으로 가장 낮은 매칭 율을 보여줬고, 제로 평균 표준화 상호 상관성 방법은 non-occlusion과 all 평가 항목에서 가장 낮 은 매칭 오차율을 보여 주지만, discontinuities 평가 항목에서는 블러 효과 때문에 높은 매칭 오차율을 보여 주었다. 시 간 복잡도 측면에서는 화소 기반인 절대 밝기차 방법이 낮은 복잡도를 보여 주였다.

Calculating of matching cost is an important for efficient stereo matching. To investigate the performance of matc hing process, the concepts of the existing methods are introduced. Also we analyze the performance and merits of th em. The simplest matching costs assume constant intensities at matching image locations. We consider matching cos t functions which can be distinguished between pixel-based and window-based approaches. The Pixel-based approac h includes absolute differences (AD) and sampling-intensitive absolute differences (BT). The window-based approac h includes the sum of the absolute differences, the sum of squared differences, the normalized cross-correlation, zero -mean normalized cross-correlation, census transform, and the absolute differences census transform (AD-Census). We evaluate matching cost functions in terms of accuracy and time complexity. In terms of the accuracy, AD-Censu s method shows the lowest matching error ratio (the best solution). The ZNCC method shows the lowest matching error ratio in non-occlusion and all evaluation part. But it performs high matching error ratio at the discontinuities e valuation part due to blurring effect in the boundary. The pixel-based AD method shows a low complexity in terms of time complexity.

As a key step in the visual inspection, image to appear particularly important. Compared with the traditional algorithm, graph cut algorithm overall high precision and faster convergence speed in discontinuous area. Based on adaptive algorithm on the basis of in-depth study, this paper proposes a three-dimensional adaptation algorithm based on graph cut theory in order to realize the image matching. Experimental results show that this algorithm can well meet the requirements of high precision and high real time capability, solve the problems such as large amount of calculation in the traditional algorithm.

Silverlight technology is a new generation of rich internet application technology. The paper proposes a new solution based on the Silverlight technology to achieve a 3D virtual reality system which takes the pictures around the objects with 360 degree displaying in the local computer or a web page. The 3D visual sense and image matching of this system are the key technical problems. The automatic matching technology in image synthesis field plays a very important role in the system. Based on the SURF algorithm, this paper takes strategies of density threshold suppression to reduce the number of matched feature points; using the quasi Euclidean distance to complete the feature points matching process, the improved algorithm also greatly reduces the matching time of feature points and improves the matching accuracy. Through theoretical analysis and experimental contrast, demonstrates the reliability and validity of the algorithm.

Stencils are finite-difference algorithms for solving large-scale and high-dimension partial differential equations. Due to the data dependences among the iterative statements in Stencils, traditional Stencil computations are be executed serially, rather than in parallel. It’s challenging to design an effective and scalable Stencil parallelized method. To address the issue of 3D data space computing, we present a serial execution model based on multi-layers symmetric Stencil method and time skewing techniques. Within this model, the iteration space is divided to multiple tiles based on time skewing, where the executive process is ordered by the sequence of tiles, and the nodes in each individual tile can be swept repeatedly to improve the data locality. In addition, we propose a novel 3D iterative space alternate tiling Stencil parallel method, which subdivides the iteration space along high dimension, and changes the execution sequence of tiles to reduce the data dependency and communication cost, where the partial order of tiles is still guaranteed. Experimental results demonstrate our proposed alternative tiling parallel method achieves better parallel efficiency and scalability compared with the domain-decomposition methods.

Block-matching and 3D filtering (BM3D) denoising algorithm [1] proposed recently has a problem of computational burden especially for low noise level and a sharp performance drop for high noise level. To solve it, an improved version of BM3D is proposed. The solution combines the digital image characteristic with added noise pollution levels, and adaptively selects block-matching threshold in grouping stage. Experimental results demonstrate it outperforms not only in terms of objective criteria of PSNR and running time, but also in visual quality.

According to mass point clouds without explicit topology relation, a parallel fast sort algorithm is proposed in this paper. Morton order is introduced and used to merge one-dimensional data. The mass point clouds of irregular model are generated corresponding address code named Morton code and these points are stored in the octree structure chain. And then a parallel fast sort algorithm based on Euclidean distance is used to sort by CPU and GPU. The k-Nearest Neighbors of point can be located in the chain. The experiment results show that much time is saved and k-Nearest Neighbors of point can be searched directly. This algorithm is simpler than those complex sort methods used on the whole point clouds.

According to the limitations of a single measurement algorithm in the current 3D models’ viewpoint extraction, this essay puts forward a viewpoint extraction algorithm based on AdaBoost iterative algorithm, which can make the features adaptive automatically. It, firstly, extracts 3D models’ feature descriptor and feature vector in the model library and adopts AdaBoost iterative algorithm to establish rules about classification and matching from geometric features and various viewpoint extraction algorithm; then, it constructs decision classifier in order to extract optimal viewpoint. In query process, the model obtains viewpoint extraction algorithm which can suit its geometric feature through decision classifier and then gets its best view by calculation. The experimental result shows this algorithm extraction effect is superior to the one by a single measurement algorithm.