• 대한전기학회논문지:시스템및제어부문D
• /
• 제50권10호
• /
• pp.473-478
• /
• 2001

Image processings using the two-dimensional wavelet transform (2DWT) have been a very active research area in recent years because the 2DWT possess many good properties. However, the discrete 2DWT can not be used for pattern recognition directly because it does not have the translation property. In this paper, we show why conventional discrete two-dimensional wavelet transforms cannot be used for pattern recognitions directly. Then, we propose a new method that makes it possible to use discrete 2DWT to pattern recognition without modification of standard pyramidal algorithms. The main idea of our method is to postprocess the wavelet transformed images using the second-order neural network. To justify the validity of the method, evaluations with test images were performed. The effectiveness of the method can be shown by the evaluation results.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송공학회 2009년도 IWAIT
• /
• pp.446-449
• /
• 2009

In this paper, we present a novel method for reconstructing a super-resolution image using multi-view low-resolution images captured for depth varying scene without requiring complex analysis such as depth estimation and feature matching. The proposed method is based on the iterative back projection technique that is extended to the 3D volume domain (i.e., space + depth), unlike the conventional superresolution methods that handle only 2D translation among captured images.

• Journal of Electrical Engineering and Technology
• /
• 제9권1호
• /
• pp.363-371
• /
• 2014

In this paper, a new super-resolution algorithm is proposed by using successive frames for generating high-resolution frames with better quality than those generated by other conventional interpolation methods. Generally, each frame used for super-resolution must only have global translation and motions of sub-pixel unit to generate good result. However, the newly proposed MSR algorithm in this paper is exempt from such constraints. The proposed algorithm consists of three main processes; motion estimation for image registration, normalization of motion vectors, and pattern analysis of edges. The experimental results show that the proposed algorithm has better performance than other conventional algorithms.

• 한국전기전자재료학회:학술대회논문집
• /
• 한국전기전자재료학회 2003년도 춘계학술대회 논문집 기술교육전문연구회
• /
• pp.109-114
• /
• 2003

Depth error correction effect for maladjusted stereo cameras with calibrated pixel distance parameter is presented. The camera calibration is a necessary procedure for stereo vision-based depth computation. Intra and extra parameters should be obtain to determine the relation between image and world coordination through experiment. One difficulty is in camera alignment for parallel installation: placing two CCD arrays in a plane. No effective methods for such alignment have been presented before. Some amount of depth error caused from such non-parallel installation of cameras is inevitable. If the pixel distance parameter which is one of intra parameter is calibrated with known points, such error can be compensated in some amount. Such error compensation effect with the calibrated pixel distance parameter is demonstrated with various experimental results.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2022년도 추계학술대회
• /
• pp.123-125
• /
• 2022

본 논문에서는 판별자를 활용하여 Image to Image translation(I2I) 분야에서 사용되는 적대적 생성 신경망(GAN)을 압축하는 방법을 제시한다. 우선, 잘 학습된 판별자와 생성자 사이의 adversarial loss 를 활용하여 생성자 내 필터들의 중요도 점수를 매겨준다. 그리고 생성자 내의 필터들을 중요도 점수를 기준으로 나열한 후 점수가 낮은 필터들을 제거하는 필터 프루닝을 한번 수행하여 적은 시간 비용으로 생성자를 압축한다. 마지막으로 지식 증류를 활용해 압축된 생성자를 학습시켜 기존의 생성자와 유사한 성능을 보이도록 하였다. 이 과정들을 통해 효과적이고 빠르게 GAN 모델을 압축할 수 있음을 확인하였다.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2021년도 추계학술대회
• /
• pp.280-282
• /
• 2021

최근에 딥러닝 기술인 GAN (Generative Adversarial Networks) 연구는 Image-to-Image translation 분야에서 활발하게 이뤄지고 있다. 이러한 기술을 바탕으로 사용자에게 편의와 재미를 제공하는 서비스가 애플리케이션 및 웹사이트의 형태로 개발되고 있다. 이에 본 논문은 CycleGAN 모델을 사용하여 이미지를 변환하고, 이를 인터랙티브 웹페이지를 통해 사용자와 실시간으로 상호작용하며 결과 이미지를 제공할 수 있는 방법을 연구하였다. 모델을 구현하기 위해 Tensorflow 및 Keras를 사용하였고, Django와 HTML5, CSS, JavaScript를 사용하여 웹사이트를 제작하였다.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2000년도 제15차 학술회의논문집
• /
• pp.407-407
• /
• 2000

For acquisition of 3-Dimensional information in real space, stereo vision system is suitable. In the stereo system, 3D real world position is derived from translation of coordinates between cameras and world. Thus, to use stereo vision, it is needed to construct a precise system which provides kinematically precise translation between camera and world coordinate, in spite of intricacy and hardness. So much cost and time should be spent to build the system. In this paper, facilely to solve previous problem, a method which can easily obtain 3D informations using reference objects and RSD(Relative Stereo Disparity) is proposed. Instead of direct computation of position with translation of coordinates, only relative stereo disparity in stereo pair of image is used to find the reference depth of objects, and real 3D position is computed with initial condition of reference objects. In computation, subpixel resolution is involved to find the display for accuracy. To find the RSD, corresponding points are calculated in subpixel resolution. So the result in experiemnt will be shown that subpixel resolution is more accurate than 1 pixel resolution.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2000년도 제13회 신호처리 합동 학술대회 논문집
• /
• pp.901-904
• /
• 2000

A direct authentication from gray-scale image, instead of the conventional multi-step preprocessing, is proposed using Gabor filter-based features from the gray-scale fingerprint around core point. The core point is located as a reference point for the translation invariant matching. And its principal symmetry axis is detected for the rotation invariant matching from its neighboring region centered at the core point. And then fingerprint is divided into non-overlapping blocks with respect to the core point and features are directly extracted form the blocked gray level fingerprint using Gabor filter. The proposed fingerprint authentication is based on the Euclidean distance between the corresponding Gabor features of the input and the template fingerprints. Experiments are conducted on 300${\times}$300 fingerprints obtained from a CMOS sensor with 500 dpi resolution, and the proposed method could lower the False Reject Rate(FRR) to 18.2% under False Acceptance Rate(FAR) of 0%.

• Journal of the Korean Academy of Child and Adolescent Psychiatry
• /
• 제29권3호
• /
• pp.122-128
• /
• 2018

Objectives: This study evaluated the psychometric properties of the Korean version of the attention-deficit/hyperactivity disorder (ADHD) Stigma Questionnaire (ASQ) and the effect of the source of information about mental health on ADHD stigma. Methods: The Korean translation of the ASQ was prepared, and 673 participants, 20-64 years of age, completed the questionnaire using an online panel survey in South Korea. The participants also completed questionnaires detailing sociodemographic variables and the source of their mental health knowledge. Cronbach's alpha coefficient was used to explore the internal consistency of the ASQ. Factor analysis using Varimax rotation was conducted to investigate the structure of the ASQ. Results: The 26-item ASQ demonstrated excellent internal consistency (Cronbach's alpha=0.940). Factor analysis supported a three-factor structure, including Concerns with Public Attitudes, Negative Self-Image, and Disclosure Concerns. There were no significant differences in the total ASQ scores according to sociodemographic characteristics. Participants who reported the internet as their major source of information about mental health showed higher ASQ scores compared to those who used other sources for mental health information. Conclusion: The Korean translation of the ASQ has acceptable psychometric properties among Korean adults. Inaccurate information from the internet could increase the stigma toward ADHD.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2019년도 춘계학술발표대회
• /
• pp.451-454
• /
• 2019

In this paper, we propose a verifiable image transformation networks to transform face sketch to photo and vice versa. Face sketch-photo is very popular in computer vision applications. It has been used in some specific official departments such as law enforcement and digital entertainment. There are several existing face sketch-photo synthesizing methods that use feed-forward convolution neural networks; however, it is hard to assure whether the results of the methods are well mapped by depending only on loss values or accuracy results alone. In our approach, we use two Resnet encoder-decoder networks as image transformation networks. One is for sketch-photo and another is for photo-sketch. They depend on each other to verify their output results during training. For example, using photo-sketch transformation networks to verify the photo result of sketch-photo by inputting the result to the photo-sketch transformation networks and find loss between the reversed transformed result with ground-truth sketch. Likely, we can verify the sketch result as well in a reverse way. Our networks contain two loss functions such as sketch-photo loss and photo-sketch loss for the basic transformation stages and the other two-loss functions such as sketch-photo verification loss and photo-sketch verification loss for the verification stages. Our experiment results on CUFS dataset achieve reasonable results compared with the state-of-the-art approaches.