• 전자공학회논문지
• /
• 제54권3호
• /
• pp.55-62
• /
• 2017

비음수 행렬 분해 기법(non-negative matrix factorization)은 대표적인 부분 영역 기반 표현 기법의 하나로 영상의 부분적인 특징을 나타내는 기저 벡터의 선형 조합으로 영상을 표현하는 기법이다. 본 논문에서는 여러 가지 비음수 행렬 분해 기법을 이용하여 얼굴 영상을 표현하고, 추출된 특징을 기반으로 학습 벡터 양자화를 이용하여 얼굴 인식을 수행하였다. 추출된 각 기법의 기저 벡터를 비교하여 각 기법의 특징을 분석하였다. 또한 NMF 기법들의 인식율 검증을 통해 비음수 행렬 기법의 얼굴 인식에 대한 활용 가능성을 확인하였다.

• Design & Manufacturing
• /
• 제2권4호
• /
• pp.37-43
• /
• 2008

Supplementary features in injection molded products, which are boss, rib and snap fit, are mainly located in the products. These features might make molding flow improper in injection processing and consequently give rise to some of molding troubles such as short shot and hesitation. The sink mark on boss parts is generated by the volumetric shrinkage that is caused by both the molding thickness and the closed boss height. The volumetric shrinkage is affected by packing pressure and its amount tends to decrease by increasing the packing pressure. The packing pressure can therefore increase flow rate to a boss part and causes the sink mark depth to increase. As the molding thickness and the closed boss height in the boss part can increase the part volume, these may yield bad solidifying and also extend the molding cycle. In this paper, both the injection molding test and the flow analysis were carried out to investigate the effect of sink marks generated in the boss part of injection molded products.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제31권10호
• /
• pp.1412-1417
• /
• 2004

전체관적인 표현방법인 희소 코딩 또는 독릴 성분 분해(ICA)는 이전의 청각의 처리와 소리 분류의 작업을 해명하는데 성공적으로 적용되었다. 반대로 부분 기반 표현법은 뇌에서 물체를 인식하는 방법을 이해하는 또 다른 방법이다. 이 논문에서, 우리는 소리 분류의 작업에 부분기반 표현법을 학습시키는 비음수화 행렬 분해(NMF)(1) 방법을 적용하였다. 잡음이 존재할 때와 존재하지 않을 때 두 가지 상황에서, NMF를 이용하여 주파수-시간영역의 소리로부터 특징을 추출하는 방법을 설명한다. 실험결과에서는 NMF에 기반을 둔 특징이 ICA에 기반을 두어 추출한 특징보다 소리 분류의 성능을 향상시킴을 보여준다.

• 한국컴퓨터그래픽스학회논문지
• /
• 제27권3호
• /
• pp.65-74
• /
• 2021

3차원 형상 복원(3D reconstruction)은 이미지 또는 영상 속 물체를 3차원 형상으로 복원하는 것을 말한다. 본 연구는 물체의 전반적 형상을 넘어 세부적인 모습까지 복원할 수 있는 표현력을 가진 3차원 형상 복원 네트워크인, 점진적 점유 네트워크를 제안한다. 본 연구가 제안하는 네트워크는 이미지 전체의 정보를 담고 있는 특징(feature)을 사용하는 기존 점유 네트워크와 달리, 수용 영역(receptive field)의 크기에 따라 다양한 수준의 이미지 특징을 추출해서 사용한다. 그리고, 다양한 수준의 이미지 특징을 디코더(decoder) 내 디코더 블록(decoder block)들에 순차적으로 반영하여, 형상 복원의 품질이 단계적으로 개선하는 네트워크 구조를 제안한다. 본 연구는 또한, 다양한 수준의 이미지 특징을 적절히 조합하여 사용하는 디코더 블록구조를 제안한다. 본 연구는 제안하는 네트워크의 성능 검증을 위해 ShapeNet 데이터 세트를 사용하였으며, 기존의 점유 네트워크(ONet) 및 다양한 수준의 이미지 특징을 사용하는 최신 연구(DISN)와 성능 비교하였다. 그 결과, 기존 점유 네트워크 대비 세 가지 검증 지표 모두에서 높은 성능을 달성하였으며, DISN과는 대등한 수준의 성능을 보여주었다. 그리고 복원 형상의 시각적 비교 결과, 본 연구의 점진적 점유 네트워크가 기존 점유 네트워크 대비, 물체의 세부 모습을 잘 복원하는 것을 확인하였다. 또한, DISN이 복원 실패한 물체의 얇은 부분 또는 이미지에서 가려진 부분을 본 연구의 네트워크는 잘 잡아내는 결과를 확인할 수 있었다. 이러한 결과는 본 연구가 제안하는 점진적 점유 네트워크의 유용성을 검증하는 결과다.

• Journal of Ginseng Research
• /
• 제42권3호
• /
• pp.334-342
• /
• 2018

Background: Panax notoginseng is a highly valued medicine and functional food, whose quality is considered to be influenced by the size, botanical parts, and growth environments. Methods: In this study, a HPLC method integrating fingerprinting and determination of multiconstituents by single reference standard was established and adopted to investigate the chemical profiles and active constituent contents of 215 notoginseng samples with different sizes, from different botanical parts and geographical regions. Results: Chemical differences among main root, branch root, and rotten root were not distinct, while rhizome and fibrous root could be discriminated from other parts. The notoginseng samples from Wenshan Autonomous Prefecture and cities nearby were similar, whereas samples from cities far away were not. The contents of major active constituents in main root did not correlate with the market price. Conclusion: This study provided comprehensive chemical evidence for the rational usage of different parts, sizes, and growth regions of notoginseng in practice.

• International Journal of Precision Engineering and Manufacturing
• /
• 제1권1호
• /
• pp.133-145
• /
• 2000

Plastic molding designers are frequently faced with optimizing multiple defects in injection molded parts. these defects are usually in conflict with each other, and thus a tradeoff needs to be made reach a final compromised solution. In this study, an automated injection molding design methodology has been developed to optimize multiple defects of injection molded parts. Two features of the proposed methodology are as follows: one is to apply the utility theory to transform the original multiple objective optimization problem into single objective optimization problem with utility as objective function, the other is an implementation of a direct search-based injection molding optimization procedure with automated consideration of process variation. The modified complex method is used as a general optimization tool in this research. The developed methodology was applied to an actual molding design and the results showed that the methodology was useful through the CAE simulation using a commercial injection molding software package. Applied to production, this study will be of immense value to industry in reducing the product development time and enhancing the product quality.

• 한국정밀공학회지
• /
• 제17권9호
• /
• pp.210-221
• /
• 2000

Plastic mold designers and frequently faced with optimizing multi-quality issues in injection molded parts. These issues are usually in conflict with each other and thus tradeoff needs to be made to reach a final compromised solution. in this study an automated injection molding design methodology has been developed to optimize multi-quality characteristics of injection molded parts. The features of the proposed methodology are as follows : first utility theory is applied to transform the original multi-objective problem into single-objective problem. Second is an implementation of a direct search-based injection molding optimization procedure with automated consideration of robustness against process variation. The modified complex method is used as a general optimization tool in this study. The developed methodology was applied to an actual mold design and the results showed that the methodology was useful through the CAE simulation using a commercial injection molding software package. Applied to production this study will be of immense value to companies in reducing the product development time and enhancing the product quality.

• 복식
• /
• 제63권8호
• /
• pp.27-42
• /
• 2013

The study was aimed to develop men's working pants patterns through the computerized 3-D virtual clothing simulation system and to verify the effects of the 3-D simulated outfit by comparing it to the images of the actual outfits. The average body measurements of South Korean men aged between 30 and 39 used for the simulation in order to generate a 3-D virtual model and to realize outfits of men's working pants for the workers in the heavy industry in South Korea. And also the preliminary questionnaire survey results on certain aspects of the working pants such as type, detailed design preference and discomforting parts were taken into consideration. Both the simulated and real images of the developed working pants were compared in terms of the ease amount according to parts of the working pants, the position of seam lines, the appearance of darts and pleats, and the effects of the fabric surface according to expertise panels' subjective 5-point scale evaluation. The results throughout the study were (1) the basic working pants item worn by subject workers were the straight one pleated pants. The most discomforting parts of the working pants were in the order of body rise, thigh, hip, waist, pants hems and knee girth. (2) the drafting factors of pants patterns differed by the men's body features, which was related to the allocation of suppression amounts between waist and hip girths into darts and hip curve amounts on the waist line level of the pants. (3) the similarity of the virtually simulated and real images of men's working pants resulted in an average of 4.5 to the ease of appearance, 4.6 to the seam lines, 4.1 to the fabric surface effects in a 5-point scale, which means that the two were highly alike.

• 한국정보기술학회논문지
• /
• 제16권11호
• /
• pp.105-113
• /
• 2018

본 논문에서는 사람 얼굴의 눈, 코, 입을 효과적으로 분류하는 방법을 제안한다. 최근 대부분의 이미지 분류는 CNN(Convolutional Neural Network)을 이용한다. 그러나 CNN으로 추출한 특징은 충분하지 않아 분류 효과가 낮은 경우가 있다. 분류 효과를 더 높이기 위해 새로운 알고리즘을 제안한다. 제안하는 방법은 크게 세 부분으로 나눌 수 있다. 첫 번째는 Haar 특징추출 알고리즘을 사용하여 얼굴의 눈, 코, 입 데이터?을 구성한다. 두번째는 CNN 구조 중 하나인 AlexNet을 사용하여 이미지의 CNN 특징을 추출한다. 마지막으로 Haar 특징 추출 뒤에 합성(Convolution) 연산을 수행하여 Haar-CNN 특징을 추출한다. 그 후 CNN 특징과 Haar-CNN을 혼합하여 Softmax를 이용해 분류한다. 혼합한 특징을 사용한 인식률은 기존의 CNN 특징 보다 약 4% 향상되었다. 실험을 통해 제안하는 방법의 성능을 증명하였다.

• 산업경영시스템학회지
• /
• 제40권2호
• /
• pp.145-149
• /
• 2017

One of the challenges facing precision manufacturers is the increasing feature complexity of tight tolerance parts. All engineering drawings must account for the size, form, orientation, and location of all features to ensure manufacturability, measurability, and design intent. Geometric controls per ASME Y14.5 are typically applied to specify dimensional tolerances on engineering drawings and define size, form, orientation, and location of features. Many engineering drawings lack the necessary geometric dimensioning and tolerancing to allow for timely and accurate inspection and verification. Plus-minus tolerancing is typically ambiguous and requires extra time by engineering, programming, machining, and inspection functions to debate and agree on a single conclusion. Complex geometry can result in long inspection and verification times and put even the most sophisticated measurement equipment and processes to the test. In addition, design, manufacturing and quality engineers are often frustrated by communication errors over these features. However, an approach called profile tolerancing offers optimal definition of design intent by explicitly defining uniform boundaries around the physical geometry. It is an efficient and effective method for measurement and quality control. There are several advantages for product designers who use position and profile tolerancing instead of linear dimensioning. When design intent is conveyed unambiguously, manufacturers don't have to field multiple question from suppliers as they design and build a process for manufacturing and inspection. Profile tolerancing, when it is applied correctly, provides manufacturing and inspection functions with unambiguously defined tolerancing. Those data are manufacturable and measurable. Customers can see cost and lead time reductions with parts that consistently meet the design intent. Components can function properly-eliminating costly rework, redesign, and missed market opportunities. However a supplier that is poised to embrace profile tolerancing will no doubt run into resistance from those who would prefer the way things have always been done. It is not just internal naysayers, but also suppliers that might fight the change. In addition, the investment for suppliers can be steep in terms of training, equipment, and software.