• Proceedings of the Korean Society of Computer Information Conference
• /
• 2019.07a
• /
• pp.255-256
• /
• 2019

한글 폰트를 만드는 데는 자음+모음 조합으로 약 11,500자 정도의 글자가 필요하다. 디자이너가 글자 하나씩 전부 디자인 하는 것도 굉장한 부담요소이고, 한글폰트를 제작하는데 있어 3개월 이상의 소요 기간과 3000만 원 이상의 비용부담 또한 무시 못 할 요소이다. 게다가 카피라이트 폰트에 대한 저작권 문제 또한 골칫거리다. 그래서 이를 최소한으로 하고자 딥 러닝의 방식중 하나인 GAN(생성적 적대 신경망)을 통해서 디자이너가 399자만 작성하고 나머지는 컴퓨터가 디자이너의 폰트 디자인을 인식하고 자동으로 만들어 주는 프로그램을 고안하였다.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.31 no.7 s.262
• /
• pp.739-746
• /
• 2007

Recently, models in STL files are widely used in reverse engineering processes, CAD systems and analysis systems. However the models have poor geometric quality and include only triangles, so the models are not suitable for the finite element analysis. This paper presents a general method that generates finite element mesh from STL file models. Given triangular meshes, the method estimates triangles and makes clusters which consist of triangles. The clusters are merged by some geometric indices. After merging clusters, the method applies plane meshing algorithm, based on domain decomposition method, to each cluster and then the result plane mesh is projected into the original triangular set. Because the algorithm uses general methods to generate plane mesh, we can obtain both tri and quad meshes unlike previous researches. Some mechanical part models are used to show the validity of the proposed method.

• Korean Journal of Computational Design and Engineering
• /
• v.11 no.2
• /
• pp.138-147
• /
• 2006

A systematic approach to tetrahedral element or mesh generation, based on an advancing-front method and an optimal triangular mesh generation technique on the surface, is presented in this paper. The possible internal nodes are obtained by the octree-decomposition scheme. The initial tetrahedral mesh system is constructed by the advancing-front method and then it is improved by the quality improvement processes including mesh swapping and nodal smoothing. The approach is evaluated by investigating the normalized length, the normalized volume, the dihedral angle and the normalized quality

• Korean Journal of Computational Design and Engineering
• /
• v.5 no.1
• /
• pp.95-101
• /
• 2000

For last two decades numerous automatic mesh generation algorithms for various two dimensional objects have been introduced continuously and among them triangular mesh generation schemes have been majority because of efficiency and controllability. In our study, an existing triangular mesh generation algorithm developed by Lo is totally modified to more improve node distribution, element shape, and objects shape independency. ft is composed of node generation part and element generation part. In order to find a suitable node position within geometry, the suggested algorithm searches desirable positions of points within boundary and optimizes node position to generate comparatively well-shaped elements. More over, the suggested algorithm handles various complex two dimensional objects and its meshing speed shows superiority to those of the existing triangulation mesh generation algorithms. It is fully automated in a sense of constructing object boundary and hence can be directly used as an independent meshing software.

• Journal of the Korean Society for Precision Engineering
• /
• v.20 no.6
• /
• pp.189-196
• /
• 2003

The NURBS surfaces are widely employed for exchanging geometric models between different CAD/CAE systems. However if the input NURBS surfaces are poorly parameterized, most surface meshing algorithms may fail or the constructed meshes can be ill-conditioned. In this paper presents a new method is presented that can generate well conditioned meshes even on poorly parameterized NURBS surfaces by regenerating NURBS surfaces. To begin with, adequate points are sampled on original poorly parameterized surfaces and new surfaces are created by interpolating these points. And then, mesh generation is performed on new surfaces. With this method, models with poorly parameterized NURBS surfaces can be meshed successfully.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2006.05a
• /
• pp.423-424
• /
• 2006

This paper proposes a hexahedral mesh generation scheme based on mapping approach and improves the drawback of sweeping algorithm. In order to improve the drawback, the algorithm in this paper generates hexahedral meshes by three dimensional element mapping first. Then hexahedral meshes are equivalent to geometry of the volume by mapping and smoothing. Sample meshes are constructed to demonstrate the mesh generating capability of the proposed algorithm.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2002.10a
• /
• pp.784-787
• /
• 2002

The NURBS surfaces are widely employed for exchanging geometric models between different CAD/CAE systems. However if the input NURBS surfaces are poorly parameterized, most surface meshing algorithms may fail or the constructed meshes can be ill-conditioned. In this paper presents a new method is presented that can generate well conditioned meshes even on poorly parameterized NURBS surfaces by regenerating NURBS surfaces. To begin with, adequate Points are sample on original poorly parameterized surfaces and new surfaces are created by interpolating these points. And then. mesh generation is performed on new surfaces. With this method, models with poorly parameterized NURBS surfaces can be meshed successfully.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2005.06a
• /
• pp.844-847
• /
• 2005

A seamless analysis of complex geometry is one of greatly interesting topic. However, there are still gaps between the industrial applications and fundamental academic studies owing to time consuming modeling process. To resolve this problem, an auto mesh generation program based on grid-based approach has been developed for IT-product in the present study. At first, base mesh and skin mesh are generated using the information of entities which extracted from IGES file. Secondly the provisional core mesh with rugged boundary geometry is constructed by superimposing the skin mesh as well as the base mesh generated from the CAD model. Finally, the positions of boundary nodes are adjusted to make a qualified mesh by adapting node modification and smoothing techniques. Also, for the sake of verification of mesh quality, the hexahedral auto mesh constructed by the program is compared with the corresponding tetrahedral free mesh and hexahedral mapped mesh through static finite element analyses. Thereby, it is anticipated that the grid-based approach can be used as a promising pre-processor for integrity evaluation of various IT-products.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.10a
• /
• pp.267-273
• /
• 1998

셀룰라 오토마타 신경망은 저자에 의하여 개발된 신경망으로써 주변의 셀과 국소적인 연결을 가지며 셀룰라 오토마타의 발생규칙에 따라 생성되는 신경망이다. 셀룰라 오토마타 신경망을 간단히 줄여서 ECANS라고 한다. 본 신경망은 카오스 뉴런 모델을 사용하며 뉴런사이의 연결강도는 흥분성 또는 억제성 결합을 갖는다. 신호의 전달방식은 펄스의 형태로서 뉴런이 발화하면 '1' 발화하지 않으면 '0'이 된다. 본 논문에서는 셀룰라 오토마타를 구성하는 요소별 특징을 살펴보고 주어진 문제에 적합한 셀룰라 오토마타 신경망을 얻어내기 위한 진화방법으로서 DNA 코딩방법을 제안한다. 제안한 방법의 유효성을 시뮬레이션을 통하여 검증한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.213-216
• /
• 2019

영상 잡음 제거 알고리즘은 잡음으로 오염된 영상으로부터 잡음이 제거된 깨끗한 영상을 추정하여 복원하는 연구이다. 기존의 모델 기반 방법의 영상 잡음 제거 알고리즘은 영상을 복원하는 과정에서 최적화 문제를 풀어야 한다는 단점과 매개변수를 직접 선택을 해주어야 한다는 단점을 가진다. 본 논문에서는 딥러닝을 이용한 학습기반 방법의 영상 잡음 제거 연구를 소개한다. 먼저, 신경망의 구축을 위하여 신경망의 구성 요소는 Instance Normalization 과 컨볼루션 신경망을 이용한 모델을 제안하였고, 여러 연구 분야에서 좋은 성능을 보이는 U-Net 구조를 전체적인 구조로 차용하였다. 신경망의 학습을 위하여 DnCNN 에서 제안한 잡음을 학습하는 잔여 학습 기법을 채택하였고, 기존의 영상 잡음 제거 알고리즘의 단점인 결과 영상이 흐릿해지는 현상을 보완하기 위하여 생성적 적대 신경망 학습 방법을 적용하였다. 본 논문에서 제안한 신경망을 이용한 잡음 제거 영상의 결과가 기존의 연구 방법들 보다 인지적인 측면에서 좋은 결과를 보임을 확인하였다.