• Journal of Korea Multimedia Society
• /
• v.16 no.4
• /
• pp.483-492
• /
• 2013

SURF is an algorithm which extracts feature points and generates their descriptors from input images, and it is being used for many applications such as object recognition, tracking, and constructing panorama pictures. Although SURF is known to be robust to changes of scale, rotation, and view points, it is hard to implement it in real time due to its complex and repetitive computations. Using 3.3 GHz Pentium, in our experiment, it takes 240ms to extract feature points and create descriptors in a VGA image containing about 1,000 feature points, which means that software implementation cannot meet the real time requirement, especially in embedded systems. In this paper, we present a hardware architecture that can compute the SURF algorithm very fast while consuming minimum hardware resources. Two key concepts of our architecture are parallelism (for repetitive computations) and efficient line memory usage (obtained by analyzing memory access patterns). As a result of FPGA synthesis using Xilinx Virtex5LX330, it occupies 101,348 LUTs and 1,367 KB on-chip memory, giving performance of 30 frames per second at 100 MHz clock.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.17 no.6
• /
• pp.1412-1422
• /
• 1993

As a part of development of process planning system for mold die manufaturing, a software system is developed, which recognizes features and extracts parameters of the shape from design data produced by solid modeller. The recognized feature date is fed to process planning and operation planning system. Low level geometry and topology data from commercial CAD system is transformed to high level machining feature data which used to be done by using a dedicated design system. The recognition algorithm is applied to the design data with boundary representation produced by a core modeller ACIS which has object oriented open architecture and is expected to become a common core modeller of next generation CAD system. The algoritm of recognition has been formulated for 21 features on prismatic components, but the feature set can be expanded by adding rules for the additional features.

• Journal of Digital Convergence
• /
• v.11 no.9
• /
• pp.311-318
• /
• 2013

For face detection techniques, the correctness of detection decreases with different lightings and backgrounds so such requires new methods and techniques. This study has aimed to obtain data for reasoning human emotional information by analyzing the components of the eyes and mouth that are critical in expressing emotions. To do this, existing problems in detecting face are addressed and a detection method that has a high detection rate and fast processing speed good at detecting environmental elements is proposed. This method must detect a specific part (eyes and a mouth) by using Haar-like Feature technique with the application of an integral image. After which, binaries detect elements based on color information, dividing the face zone and skin zone. To generate correct shape, the shape of detected elements is generated by using a bezier curve-a curve generation algorithm. To evaluate the performance of the proposed method, an experiment was conducted by using data in the Face Recognition Homepage. The result showed that Haar-like technique and bezier curve method were able to detect face elements more elaborately.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.27 no.2
• /
• pp.195-202
• /
• 2009

As the generation and application of spatial information is gradually expanded not only in traditional surveying fields but also a CNS and an ITS recently. The Accuracy and the newest of data grow to be an important element. But digital map is updated with system based tile. So, it is hard to get the newest of data and to be satisfied with user requirements. In this study, management system is developed to manage feature change efficiently using bid informations from NaraJangter which service the bid informations. A construction works with change possibility of feature from bid informations are classified and are made DB. And the DB is used as the feature change forecast informations. Also, It is converted from bid information of text form to positioning informations connected to spatial information data. If this system is made successfully, this system contributes to reduce the cost for the update of digital map and to take the newest date of spatial informations.

• The Journal of the Korea Contents Association
• /
• v.18 no.11
• /
• pp.447-454
• /
• 2018

This paper proposes a depth image generation algorithm of stereo images using a deep learning model composed of a CNN (convolutional neural network). The proposed algorithm consists of a feature extraction unit which extracts the main features of each parallax image and a depth learning unit which learns the parallax information using extracted features. First, the feature extraction unit extracts a feature map for each parallax image through the Xception module and the ASPP(Atrous spatial pyramid pooling) module, which are composed of 2D CNN layers. Then, the feature map for each parallax is accumulated in 3D form according to the time difference and the depth image is estimated after passing through the depth learning unit for learning the depth estimation weight through 3D CNN. The proposed algorithm estimates the depth of object region more accurately than other algorithms.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.6
• /
• pp.2806-2825
• /
• 2018

Unsupervised learning has shown good performance on image, video and audio classification tasks, and much progress has been made so far. It studies how systems can learn to represent particular input patterns in a way that reflects the statistical structure of the overall collection of input patterns. Many promising deep learning systems are commonly trained by the greedy layerwise unsupervised learning manner. The performance of these deep learning architectures benefits from the unsupervised learning ability to disentangling the abstractions and picking out the useful features. However, the existing unsupervised learning algorithms are often difficult to train partly because of the requirement of extensive hyperparameters. The tuning of these hyperparameters is a laborious task that requires expert knowledge, rules of thumb or extensive search. In this paper, we propose a simple and effective unsupervised feature learning algorithm for image classification, which exploits an explicit optimizing way for population and lifetime sparsity. Firstly, a sparse target matrix is built by the competitive rules. Then, the sparse features are optimized by means of minimizing the Euclidean norm ($L_2$) error between the sparse target and the competitive layer outputs. Finally, a classifier is trained using the obtained sparse features. Experimental results show that the proposed method achieves good performance for image classification, and provides discriminative features that generalize well.

• Journal of the Korea Computer Industry Society
• /
• v.7 no.5
• /
• pp.545-554
• /
• 2006

RFID could be applied in the various fields such as distribution beside, circulation, traffic and environment on information communication outside. So this can speak as point of ubiquitous computing's next generation technology. However, it is discussed problem of RFID security recently, so we must prepare thoroughly about RFID security for secure information. In this paper, we proposed a method which could protect private information and ensure RFID's identification effectively storing face feature information on RFID tag. Our method which is improved linear discriminant analysis has reduced dimension of feature information which has large size of data. Therefore, we can sore face feature information in small memory field of RFID tag. Our propose d algorithm has shown 92% recognition rate in experimental results and can be applied to entrance control management system, digital identification card and others.

• Transactions of the Society of Information Storage Systems
• /
• v.1 no.1
• /
• pp.108-114
• /
• 2005

The generation of fine relics of suspensions is a significant interest as it holds the key to the fabrication of electronic devices. These processes offer opportunities for miniaturization of multilayer circuits, for production of functionally graded materials, ordered composites and far small complex-shaped components. Some novel printing methods of depositing ceramic and metal droplets were suggested in recent years. In an electro-hydrodynamic printing, the metallic capillary nozzle can be raised to several kilovolts with respect to the infinite ground plate or pin-type electrode positioned a few millimeters from the nozzle tip. Depending on the electrical and physical properties of the liquid, for a given geometry, it Is possible to generate droplets in any one of three modes, dripping, cone-jet and multi-jet. In this experiment, an alumina suspension flowing through a nozzle was subjected to electro-hydrodynamic printing using pin-type electrodes in the cone-jet mode at different applied voltages. The pin-type electrodes of 1, 100, 1000${\mu}m$ in diameter were used to form fine ceramic patterns onto the substrates. Various feature sizes with applied voltages and electrode diameters were measured. The feature sizes increased with the electrode diameter and applied voltages. The feature size was as fine as $30 {\mu}m$.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.10
• /
• pp.3230-3255
• /
• 2022

Causality mining in NLP is a significant area of interest, which benefits in many daily life applications, including decision making, business risk management, question answering, future event prediction, scenario generation, and information retrieval. Mining those causalities was a challenging and open problem for the prior non-statistical and statistical techniques using web sources that required hand-crafted linguistics patterns for feature engineering, which were subject to domain knowledge and required much human effort. Those studies overlooked implicit, ambiguous, and heterogeneous causality and focused on explicit causality mining. In contrast to statistical and non-statistical approaches, we present Bidirectional Encoder Representations from Transformers (BERT) integrated with Multi-level Feature Networks (MFN) for causality recognition, called BERT+MFN for causality recognition in noisy and informal web datasets without human-designed features. In our model, MFN consists of a three-column knowledge-oriented network (TC-KN), bi-LSTM, and Relation Network (RN) that mine causality information at the segment level. BERT captures semantic features at the word level. We perform experiments on Alternative Lexicalization (AltLexes) datasets. The experimental outcomes show that our model outperforms baseline causality and text mining techniques.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.79-81
• /
• 2022

MPEG-VCM(Video Coding for Machines)에서는 머신비전(machine vision) 네트워크의 백본(backbone)에서 추출된 이미지/비디오 특징 압축을 위한 표준화를 진행하고 있다. 현재 VCM 표준기술 탐색 과정에서 가장 좋은 압축 성능을 보이는 MSFC(Multi-Scale Feature compression) 기반 압축 네트워크 모델은 추출된 멀티-스케일 특징을 단일-스케일 특징으로 변환하여 특징맵으로 구성하고 이를 VVC 로 압축한다. 본 논문에서는 MSFC 기반 압축 모델에서 Min-Max 값 시그널링을 제외한 최소-최대(Min-Max) 정규화를 포함한 개선된 특징맵 생성 기법을 제시한다. 즉, 제안기법은 VCM 디코더에서의 특징맵 복원을 위한 Min-Max 값을 학습 기반으로 생성함으로써 Min-Max 시그널링의 비트 오버헤드 절감뿐만 아니라 별도의 시그널링 기제를 생략한 보다 단순한 전송 비트스트림 구성을 가능하게 한다. 실험결과 제안기법은 이미지 앵커(Anchor) 대비 BPP-mAP 성능에서 83.24% BD-rate 이득을 보이며, 이는 기존 MSFC 보다 1.74%정도 다소 떨어지지만 별도의 Min-Max 시그널링 없이도 기존의 성능을 유지할 수 있음을 보인다.