• Journal of Internet Computing and Services
• /
• v.4 no.6
• /
• pp.57-67
• /
• 2003

The recent rapid development of multimedia and optical technologies brings great attention to application systems to process facial Image features. The previous research efforts in facial image processing have been mainly focused on the recognition of human face and facial expression analysis, using front face images. Not much research has been carried out Into image-based detection of face direction. Moreover, the existing approaches to detect face direction, which normally use the sequential Images captured by a single camera, have limitations that the frontal image must be given first before any other images. In this paper, we propose a method to detect face direction by using facial features such as facial trapezoid which is defined by two eyes and the lower lip. Specifically, the proposed method forms a facial direction formula, which is defined with statistical data about the ratio of the right and left area in the facial trapezoid, to identify whether the face is directed toward the right or the left. The proposed method can be effectively used for automatic photo arrangement systems that will often need to set the different left or right margin of a photo according to the face direction of a person in the photo.

• 한국HCI학회:학술대회논문집
• /
• 2007.02a
• /
• pp.897-902
• /
• 2007

인간과 컴퓨터간의 전통적인 인터페이스는 인간이 요구하는 다양한 인터페이스를 제공하지 못한다는 점에서 점차 사용하기 불편하게 되었고 이는 새로운 형태의 인터페이스에 대한 요구로 이어지게 되었다. 본 논문에서는 이러한 추세에 맞추어 카메라를 통해 인간의 손 제스처를 인식하는 새로운 인터페이스를 연구하였다. 손은 자유도가 높고 3차원의 view direction에 의해 형상이 매우 심하게 변한다. 따라서 윤곽선 기반방법과 같은 2차원으로 투영된 영상에서 contour나 edge의 정보로 손 제스처를 인식하는 데는 한계가 있다. 그러나 모델기반 방법은 3차원 정보를 이용하기 때문에 손 제스처를 인식하는데 좋으나 계산량이 많아 실시간으로 처리하기가 쉽지 않다. 이러한 문제점을 해결하기 위해 손 형상에 대한 대규모 데이터베이스를 구성하고 정규화된 공간에서 Feature 간의 연관성을 파악하여 훈련 데이터 모델을 구성하여 비교함으로써 실시간으로 손 포즈를 구별할 수 있다. 이러한 통계적 학습 기반의 알고리즘은 다양한 데이터와 좋은 feature의 검출이 최적의 성능을 구현하는 것과 연관된다. 따라서 배경으로부터 노이즈를 최대한 줄이기 위해 피부의 색상 정보를 이용하여 손 후보 영역을 검출하고 검출된 후보 영역으로부터 HLF(Haar-like Feature)를 이용하여 손 영역을 검출한다. 검출된 손 영역으로부터 패턴 분류 과정을 거쳐 손 포즈를 인식 하게 된다. 패턴 분류 과정은 HLF를 이용하여 손 포즈를 인식하게 되는데 미리 학습된 각 포즈에 대한 HLF를 이용하여 손 포즈를 인식하게 된다. HLF는 Violar가 얼굴 검출에 적용한 것으로 얼굴 검출에 좋은 결과를 보여 주었으며, 이는 적분 이미지로부터 추출한 HLF를 이용한 Adaboost 학습 알고리즘을 사용하였다. 본 논문에서는 피부색의 색상 정보를 이용 배경과 손 영상을 최대한 분리하여 배경의 대부분이 Adaboost-Haar Classifier의 첫 번째 스테이지에서 제거되는 방법을 이용하여 그 성능을 더 향상 시켜 손 형상 인식에 적용하였다.

• Journal of KIISE:Computing Practices and Letters
• /
• v.8 no.5
• /
• pp.562-573
• /
• 2002

In this paper, we propose a method for measuring the volume of packages for the efficient handling of the packages. Using the geometrical characteristics of the rectangular parallelepiped type objects, the method measures the volume of packages with one camera only in real time. In preprocessing of volume measurement, the method extracts outer lines of the object and then crossing points of the lines as feature points or vertexes. From these cross points(-feature points-), the volume of the package is calculated. Compared to the direct feature extraction, the proposed method shows especially the blurring robust result by using the line for feature extraction. Additionally, the method can get the stable result by considering object's direction. From experimental results, it is demonstrated that this method is very effective for the real time volume measurement of the rectangular parallelepiped.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.12 no.3
• /
• pp.186-191
• /
• 2019

In this paper, we propose a implementation approach of the high-speed feature extraction algorithm. The proposed method is based on the block type classification algorithm which reduces the computation time when target macro block is divided to smooth block type that has no image features. It is quantitatively identified that occurs at 29.5% of the total image using 200 standard test images with $64{\times}64$ macro block size. This means that within a standard test image containing various image information, 29.5% can reduce the complexity of the operation. When the proposed approach is applied to the Canny edge detection, the required latency of the edge detection can be completely eliminated, such as 2D derivative filter, gradient magnitude/direction computation, non-maximal suppression, adaptive threshold calculation, hysteresis thresholding. Also, it is expected that operation time of the feature detection can be reduced by applying block type classification algorithm to various feature extraction algorithms in this way.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.10
• /
• pp.2768-2787
• /
• 2023

As an important research direction of the application of computer science in the medical field, the automatic generation technology of radiology report has attracted wide attention in the academic community. Because the proportion of normal regions in radiology images is much larger than that of abnormal regions, words describing diseases are often masked by other words, resulting in significant feature loss during the calculation process, which affects the quality of generated reports. In addition, the huge difference between visual features and semantic features causes traditional multi-modal fusion method to fail to generate long narrative structures consisting of multiple sentences, which are required for medical reports. To address these challenges, we propose a multi-feature optimization Transformer (MFOT) for generating radiology reports. In detail, a multi-dimensional mapping attention (MDMA) module is designed to encode the visual grid features from different dimensions to reduce the loss of primary features in the encoding process; a feature pre-fusion (FP) module is constructed to enhance the interaction ability between multi-modal features, so as to generate a reasonably structured radiology report; a detail enhanced attention (DEA) module is proposed to enhance the extraction and utilization of key features and reduce the loss of key features. In conclusion, we evaluate the performance of our proposed model against prevailing mainstream models by utilizing widely-recognized radiology report datasets, namely IU X-Ray and MIMIC-CXR. The experimental outcomes demonstrate that our model achieves SOTA performance on both datasets, compared with the base model, the average improvement of six key indicators is 19.9% and 18.0% respectively. These findings substantiate the efficacy of our model in the domain of automated radiology report generation.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.16 no.6
• /
• pp.1318-1324
• /
• 2012

This paper proposes a novel method for Game character classification based on texture and edge orientation feature. The character dose not move(NPC) and move the character is classified. Classification of property within the character of straight line segments are used to extract features. First, the character inside edge feature extraction and then calculates EEDH, SSPD. The extracted attribute represents the energy of a particular direction. Thus, these properties were used to classify of NPC and Monster. The proposed method, the user can reduce the unnecessary time in the game.

• Journal of Institute of Control, Robotics and Systems
• /
• v.10 no.1
• /
• pp.81-86
• /
• 2004

This paper describes self-localization of a mobile robot from the multiple candidates of landmarks in outdoor environment. Our robot uses omnidirectional vision system for efficient self-localization. This vision system acquires the visible information of all direction views. The robot uses feature of landmarks whose size is bigger than that of others in image such as building, sculptures, placard etc. Robot uses vertical edges and those merged regions as the feature. In our previous work, we found the problem that landmark matching is difficult when selected candidates of landmarks belonging to region of repeating the vertical edges in image. To overcome these problems, robot uses the merged region of vertical edges. If interval of vertical edges is short then robot bundles them regarding as the same region. Thus, these features are selected as candidates of landmarks. Therefore, the extracted merged region of vertical edge reduces the ambiguity of landmark matching. Robot compares with the candidates of landmark between previous and current image. Then, robot is able to find the same landmark between image sequences using the proposed feature and method. We achieved the efficient self-localization result using robust landmark matching method through the experiments implemented in our campus.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.25 no.1
• /
• pp.76-83
• /
• 2012

In a real-time indoor place recognition system using image features detection, specific markers included in input image should be detected exactly and quickly. However because the same markers in image are shown up differently depending to movement, direction and angle of camera, it is required a method to solve such problems. This paper proposes a technique to extract the features of object without regard to change of the object scale. To support real-time operation, it adopts SURF(Speeded up Robust Features) which enables fast feature detection. Another feature of this system is the user mark designation which makes possible for user to designate marks from input image for location detection in advance. Unlike to use hardware marks, the feature above has an advantage that the designated marks can be used without any manipulation to recognize location in input image.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.5
• /
• pp.653-662
• /
• 2022

Recently, various researches have been made on a fingerprint classification method using Convolutional Neural Networks (CNN), which is widely used for multidimensional and complex pattern recognition such as images. The CNN-based fingerprint classification method can be executed by integrating the two-step process, which is generally divided into feature extraction and classification steps. Therefore, since the CNN-based methods can automatically extract features of fingerprint images, they have an advantage of shortening the process. In addition, since they can learn various features of incomplete or low-quality fingerprints, they have flexibility for feature extraction in exceptional situations. In this paper, we intend to identify the research trends of CNN-based fingerprint classification and discuss future direction of research through the analysis of experimental methods and results.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.3
• /
• pp.437-447
• /
• 2023

Memory protection has been researched for decades for program execution protection. ARM recently developed a newhardware security feature to protect memory that was applied to real hardware. However, there are not many hardware withhardware memory protection feature and research has not been actively conducted yet. We perform diagnostics on howandhow it works on real hardware, and on security, with a new hardware memory protection feature, named 'Pointer Authentication Code'. Through this research, it will be possible to find out the direction, use, and security of future hardware security technologies and apply to the program.