• Journal of Broadcast Engineering
• /
• v.26 no.2
• /
• pp.125-131
• /
• 2021

In recent years, Convolutional Neural Networks (CNNs) have achieved outstanding performance in the fields of computer vision such as image classification, object detection, visual quality enhancement, etc. However, as huge amount of computation and memory are required in CNN models, there is a limitation in the application of CNN to low-power environments such as mobile or IoT devices. Therefore, the need for neural network compression to reduce the model size while keeping the task performance as much as possible has been emerging. In this paper, we propose a method to compress CNN models by combining matrix decomposition methods of LR (Low-Rank) approximation and CP (Canonical Polyadic) decomposition. Unlike conventional methods that apply one matrix decomposition method to CNN models, we selectively apply two decomposition methods depending on the layer types of CNN to enhance the compression performance. To evaluate the performance of the proposed method, we use the models for image classification such as VGG-16, RestNet50 and MobileNetV2 models. The experimental results show that the proposed method gives improved classification performance at the same range of 1.5 to 12.1 times compression ratio than the existing method that applies only the LR approximation.

• Journal of Broadcast Engineering
• /
• v.24 no.1
• /
• pp.132-141
• /
• 2019

The necessity of transmitting video data over a narrow-bandwidth exists steadily despite that video service over broadband is common. In this paper, we propose a scalable video coding framework for low-resolution video transmission over a very narrow-bandwidth network by super-resolution of decoded frames of a base layer using a convolutional neural network based super resolution technique to improve the coding efficiency by using it as a prediction for the enhancement layer. In contrast to the conventional scalable high efficiency video coding (SHVC) standard, in which upscaling is performed with a fixed filter, we propose a scalable video coding framework that replaces the existing fixed up-scaling filter by using the trained convolutional neural network for super-resolution. For this, we proposed a neural network structure with skip connection and residual learning technique and trained it according to the application scenario of the video coding framework. For the application scenario where a video whose resolution is $352{\times}288$ and frame rate is 8fps is encoded at 110kbps, the quality of the proposed scalable video coding framework is higher than that of the SHVC framework.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.15 no.2
• /
• pp.65-71
• /
• 2011

Purpose: The recently adopted multiple confocal SPECT SYSTEM (hereinafter called IQ SPECT$^{TM}$) has a high difference from the conventional myocardial perfusion SPECT in the collimator form, image capture method, and image reconstruction method. This study was conducted to compare this novice equipment with the conventional one to design a protocol meeting the IQ SPECT, and also determine the characteristics and usefulness of IQ SPECT. Materials and Methods: 1. For the objects of LEHR (Low energy high resolution) collimator and Multiple confocal collimator, $^{99m}Tc$ 37MBq was put in the acrylic dish then each sensitivity ($cpm/{\mu}Ci$) was measured at the distance of 5 cm, 10 cm, 20 cm, 30 cm, and 40 cm respectively. 2. Based on the sensitivity measure results, IQ SPECT Protocol was designed according to the conventional general myocardial SPECT, then respectively 278 kBq/mL, 7.4 kBq/mL, and 48 kBq/mL of $^{99m}Tc$ were injected into the myocardial and soft tissues and liver site by using the anthropomorphic torso phantom then the myocardial perfusion SPECT was run. 3. For the comparison of FWHMs (Full Width at Half Maximum) resulted from the image reconstruction of LEHR collimator, the FWHMs (mm) were measured with only algorithms changed, in the case of the FBP (Filtered Back projection) method- a reconstruction method of conventional myocardial perfusion SPECT, and the 3D OSEM (Ordered subsets expectation maximization) method of IQ SPECT, by using $^{99m}Tc$ Line source. Results: 1. The values of IQ SPECT collimator sensitivity ($cpm/{\mu}Ci$) were 302, 382, 655, 816, 1178, and those of LEHR collimator were measured as 204, 204, 202, 201, 198, both at the distance of 5 cm, 10 cm, 20 cm, 30 cm, and 40 cm respectively. It was found the difference of sensitivity increases up to 4 times at the distance of 30 cm in the cases of IQ SPECT and LEHR. 2. The myocardial perfusion SPECT Protocol was designed according to the geometric characteristics of IQ SPECT based on the sensitivity results, then the phantom test for the aforesaid protocol was conducted. As a result, it was found the examination time can be reduced 1/4 compared to the past. 3. In the comparison of FWHMs according to the reconstructed algorithm in the FBP method and 3D OSEM method followed after the SEPCT test using a LEHR collimator, the result was obtained that FWHM rose around twice in the 3D OSEM method. Conclusion : The IQ SPECT uses the Multiple confocal collimator for the myocardial perfusion SPECT to enhance the sensitivity and also reduces examination time and contributes to improvement of visual screen quality through the myocardial-specific geometric image capture method and image reconstruction method. Due to such benefits, it is expected patients will receive more comfortable and more accurate examinations and it is considered a further study is required using additional clinical materials.

• Journal of Broadcast Engineering
• /
• v.22 no.2
• /
• pp.193-206
• /
• 2017

In this paper, we propose a rate control method based on the $R-{\lambda}$ model that supports a parallel encoding structure in GOP levels or IDR period levels for 4K UHD input video in real-time. For this, a slice-level bit allocation method is proposed for parallel encoding instead of sequential encoding. When a rate control algorithm is applied in the GOP level or IDR period level parallelism, the information of how many bits are consumed cannot be shared among the frames belonging to a same frame level except the lowest frame level of the hierarchical B structure. Therefore, it is impossible to manage the bit budget with the existing bit allocation method. In order to solve this problem, we improve the bit allocation procedure of the conventional ones that allocate target bits sequentially according to the encoding order. That is, the proposed bit allocation strategy is to assign the target bits in GOPs first, then to distribute the assigned target bits from the lowest depth level to the highest depth level of the HEVC hierarchical B structure within each GOP. In addition, we proposed a processing method that is used to improve subjective image qualities by allocating the bits according to the coding complexities of the frames. Experimental results show that the proposed bit allocation method works well for frame-level parallel HEVC software encoders and it is confirmed that the performance of our rate controller can be improved with a more elaborate bit allocation strategy by using the preprocessing results.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.7
• /
• pp.619-628
• /
• 2000

Abstract Perspective views provide a powerful depth cue and thus aid the interpretation of complicated images. The main drawback of current perspective volume rendering is the long execution time. In this paper, we present an efficient perspective volume rendering algorithm based on coherency between rays. Two sets of templates are built for the rays cast from horizontal and vertical scanlines in the intermediate image which is parallel to one of volume faces. Each sample along a ray is calculated by interpolating neighboring voxels with the pre-computed weights in the templates. We also solve the problem of uneven sampling rate due to perspective ray divergence by building more templates for the regions far away from a viewpoint. Since our algorithm operates in object-order, it can avoid redundant access to each voxel and exploit spatial data coherency by using run-length encoded volume. Experimental results show that the use of templates and the object-order processing with run-length encoded volume provide speedups, compared to the other approaches. Additionally, the image quality of our algorithm improves by solving uneven sampling rate due to perspective ray di vergence.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.45 no.6
• /
• pp.17-26
• /
• 2008

In this paper, we propose a distributed multi-view video coding method employing illumination compensation for multi-view video coding. Distributed multi-view video coding (DMVC) methods can be classified either into a temporal or an inter-view interpolation-based ones according to ways to generate side information. DMVC with inter-view interpolation utilizes characteristics of multi-view videos to improve coding efficiency of the DMVC by using side information based on the inter-view interpolation. However, mismatch of camera parameters and illumination change between two views could bring about inaccurate side information generation. In this paper, a modified distributed multi-view coding method is presented by applying illumination compensation in generating the side information. In the proposed encoder system, in addition to parity bits for AC coefficients, DC coefficients are transmitted as well to the decoder side. By doing so, the decoder can generate more accurate side information by compensating illumination changes with the transmitted DC coefficients. We found that the proposed algorithm is $0.1{\sim}0.2\;dB$ better than the conventional algorithm that does not make use of illumination compensation.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.6
• /
• pp.55-64
• /
• 2017

High-capacity, high-definition image applications need to process considerable amounts of data at high speed. Accordingly, users of these applications demand a high-speed parallel execution system. To increase the speed of a parallel execution system, Park (2004) proposed a technique, called MAMS (Multi-Access Memory System), to access data in several execution units without the conflict of parallel processing memories. Since then, many studies on MAMS have been conducted, furthering the technique to MAMS-PP16 and MAMS-PP64, among others. As a memory architecture for parallel processing, MAMS must be constructed in one chip; therefore, a method to achieve the identical functionality as the existing MAMS while minimizing the architecture needs to be studied. This study proposes a method of miniaturizing the MAMS architecture in which the architectures of the ACR (Address Calculation and Routing) circuit and MMS (Memory Module Selection) circuit, which deliver data in memories to parallel execution units (PEs), do not use the MMS circuit, but are constructed as one shift and conditional statements whose number is the same as that of memory modules inside the ACR circuit. To verify the performance of the realized architecture, the study conducted the processing time of the proposed MAMS-PP64 through an image correlation test, the results of which demonstrated that the ratio of the image correlation from the proposed architecture was improved by 1.05 on average.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2009.05a
• /
• pp.1812-1816
• /
• 2009

이미지 해석에 의한 유속장 측정방법은 유체역학분야에서 지난 30 여년 동안 많이 활용되어온 속도측정 기법으로 오늘날에는 이를 수공학 분야에서 이를 유량측정 등 수리현상 해석에 활용하려는 시도가 다각적으로 이루어지고 있다. 이에 본 연구에서는 이미지 해석에 의한 유속장 측정방법을 용담댐 시험유역에 적용하여 그의 자연하천에서의 적용성을 검토하고자 한다. 이미지 해석에 의한 유속장 측정방법은 PIV(Particle Image Velocimetry)로 통칭되고 있으며, PIV는 seeding, illumination, recording, 및 image processing의 네 가지 요소로 구성된다. seeding을 위해서 유체를 따라 흐를수 있는 작은 입자를 유체에 첨가한다. 유체를 따라 흐르는 입자들의 선명한 이미지를 얻기 위해서illumination이 필요하다. PIV를 이용하여 흐름을 해석하기 위한 illumination은 일반적으로 이중펄스 레이저가 이용된다. 이렇게 유속장 해석을 하려는 유체에 대하여 seeding 및 illumination이 준비되면 단일노출- 다중 프레임법, 혹은 다중노출-단일 프레임법으로 흐름을 recording을 한다. image processing은 이미지를 다운로드하고, 디지타이징 및 화질향상을 하는 전처리(pre-processing), 상관계수의 산정에 의한 유속 벡터의 결정 및 에러 벡터를 제거하고 유속장을 그래프화하는 후처리(post-processing) 과정으로 구성된다. LSPIV(Large Scale PIV)는 PIV의 기본원리를 근거로 하여 기존의 PIV에 비하여 실험실 내에서의 수리모형실험이나 일반 하천에서의 유속측정과 같은 큰 규모$(4m^2\sim45,000m^2$)의 흐름해석을 할 수 있도록 Fujita et al.(1994)와 Aya et al.(1995)이 확장시킨 것이다. PIV와 비교시 LSPIV의 다른 점은 넓은 흐름 표면적을 포함하기 위하여 촬영시에 카메라의 광축과 흐름 사이의 각도가 PIV에서 이용하는 수직이 아닌 경사각을 이용하였고 이에 따라 발생하는 이미지의 왜곡을 제거하기 위하여 이미지 변환기법을 적용하여 왜곡이 없는 정사촬영 이미지로 변환시킨다. 이후부터는 PIV의 이미지 처리 방법이 적용되어 표면유속을 산정한다. 다만 이미지 변환을 PIV 이미지 처리 전에 하느냐 후에 하느냐에 따라 유속장 해석결과에 차이가 있다. PIV의 네가지 단계를 포함하여 LSPIV의 각 단계를 구분하면, seeding, illumination, recording, image transformation,image processing 및 post-processing의 여섯 단계로 나뉘어진다 (Li, 2002). LSPIV를 적용시 물표면 입자의 Tracing을 위하여 자연하천에서 사용하기에 적합한 환경친화적인 seeding 재료인 Wood Mulch를 사용하여 유속을 측정하였다. 적용지점은 용담댐 상류의 동향수위관측소 지점으로 이 지점은 한국수자원공사의 수자원시험유역이 위치하고 있다. 이미지의 촬영은 가정용 비디오 캠코더 (Sony DCR-PC 350)을 이용하여 두 줄기의 흐름에 대하여 각각 약 5분 동안의 영상을 촬영한후 이중에서 seeding의 분포가 잘 이루어진 약 1분간을 추출한후 이를 이용하여 PIV 분석에 이용하였다. 대체적으로 유속장의 계산이 무난하게 이루어지었으나 비교적 수질 상태가 양호하고, 수심이 낮고, 하상재료가 자갈로 이루어져 있어 비슷한 색상의 seeding 재료를 추적하기 어려운 구간이 발생한 부분에서는 유속의 계산이 정확히 이루어지지 않았다.

• 한국가스학회:학술대회논문집
• /
• 2007.04a
• /
• pp.195-200
• /
• 2007

본 연구는 도심지에 위치한 LPG 충전소를 연구범위로 하여 공간정보의 활용이 가장 많이 요구되는 안전관리 분야의 업무를 중심으로 공간정보를 효율적으로 구축 활용하기 위하여 데이터베이스를 중심으로 위험관리정보 시스템을 개발하였다. 이를 바탕으로 정량적 위험성 평가의 자동화를 통해 나타난 위험성을 실시간에 제어하기 위한 필요조건을 표준화하여 기초 정보자료로 구축, 이를 지리정보기능과 연동하여 LPG 충전소의 안전검사의 효율화, 사전 위험성 평가, 사고대응 판단의 효과적인 의사결정을 유도 할 수 있는 기반을 제공한다. 위험관리정보시스템(RMIS, Risk Management Information System) 개발절차는 다음과 같다 첫째, 도심지에 위치한 LPG 충전소 위험성 평가를 수행함에 있어서 기본적인 데이터인 부지내(On-site) 관련 자료와 부지 외(Off-site) 관련 자료를 관계형 데이터베이스(RDB, Relational Database)로 개발하였다. 둘째, Visual Basic을 이용하여 사용자가 효과적으로 위험을 관리 제어 할 수 있는 위험관리 통합 데이터베이스 시스템 개발하였다. 셋째, 위험관리 통합 데이터베이스 시스템과 지리정보시스템에 연동을 통한 의사결정 방안 제시하였다. 위험관리정보시스템(RMIS) 프로그램을 개발을 통하여 다음과 같은 결과를 도출하였다. 첫째, 위험관리 데이터 이용하여 사용자와 검사자가 효과적으로 위험을 사전관리 할 수 있는 공유정보를 구축하였다. 둘째, 위험 관리를 부지 내와 부지 외로 나누어 관리함으로서 시설 내부 뿐 만 아니라 시설외부에 미치는 영향을 모두 고려하여 구축하므로 서, 중대사고에 대응 할 수 있는 종합적인 안전관리 기반을 조성하였다. 셋째, 사용자 인터페이스를 바탕으로 비상사태 발생시에 신속하고 정확한 의사결정을 할 수 있는 기반을 조성하였다. 제공하여 응용GIS 구축의 생산성 및 품질 향상에 기여할 뿐만 아니라 우리의 최종목표인 GIS 소프트웨어 자동 생산에도 크게 기여할 것으로 사료된다. 등)을 교통망상에 표시할 수 있음으로서 의사결정에 보다 많은 도움을 줄 수 있을 것이다. 비트율의 증가와 화질 열화는 각각 최대 1.32%와 최대 0.11dB로 무시할 수 있을 정도로 작음을 확인 하였다.을 알 수 있었다. 현지관측에 비해 막대한 비용과 시간을 절약할 수 있는 위성영상해석방법을 이용한 방법은 해양수질파악이 가능할 것으로 판단되며, GIS를 이용하여 다양하고 복잡한 자료를 데이터베이스화함으로써 가시화하고, 이를 기초로 공간분석을 실시함으로써 환경요소별 공간분포에 대한 파악을 통해 수치모형실험을 이용한 각종 환경영향의 평가 및 예측을 위한 기초자료로 이용이 가능할 것으로 사료된다.염총량관리 기본계획 시 구축된 모형 매개변수를 바탕으로 분석을 수행하였다. 일차오차분석을 이용하여 수리매개변수와 수질매개변수의 수질항목별 상대적 기여도를 파악해 본 결과, 수리매개변수는 DO, BOD, 유기질소, 유기인 모든 항목에 일정 정도의 상대적 기여도를 가지고 있는 것을 알 수 있었다. 이로부터 수질 모형의 적용 시 수리 매개변수 또한 수질 매개변수의 추정 시와 같이 보다 세심한 주의를 기울여 추정할 필요가 있을 것으로 판단된다.변화와 기흉 발생과의 인과관계를 확인하고 좀 더 구체화하기 위한 연구가 필요할 것이다.게 이루어질 수 있을 것으로 기대된다.는 초과수익률이 상승하지만, 이후로는 감소하므로, 반전거래전략을 활용하는 경우 주식투자기간은 24개월이하의 중단기가 적합함을 발견하였다. 이상의 행태적 측면과 투자성과측면의 실증결과를 통하여 한국주식시장에 있어서 시장수익률을 평균적으로 초과할 수 있는 거래전략은 존재하므로 이러한 전략을 개발 및 활용할 수 있으며, 특히, 한국주

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.4 s.310
• /
• pp.37-45
• /
• 2006

In this paper, we propose the development of MHLA(Modulatory Hippocampus Learning Algorithm) which remodel a principle of brain of hippocampus. Hippocampus takes charge auto-associative memory and controlling functions of long-term or short-term memory strengthening. We organize auto-associative memory based 3 steps system(DG, CA3, CAl) and improve speed of learning by addition of modulator to long-term memory learning. In hippocampal system, according to the 3 steps order, information applies statistical deviation on Dentate Gyrus region and is labelled to responsive pattern by adjustment of a good impression. In CA3 region, pattern is reorganized by auto-associative memory. In CAI region, convergence of connection weight which is used long-term memory is learned fast by neural networks which is applied modulator. To measure performance of MHLA, PCA(Principal Component Analysis) is applied to face images which are classified by pose, expression and picture quality. Next, we calculate feature vectors and learn by MHLA. Finally, we confirm cognitive rate. The results of experiments, we can compare a proposed method of other methods, and we can confirm that the proposed method is superior to the existing method.