• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.3
• /
• pp.112-118
• /
• 2009

Skin color segmentation techniques have been widely utilized for face/hand detection and tracking in many applications such as a diagnosis system using facial information, human-robot interaction, an image retrieval system. In case of a video image, it is common that the skin color model for a target is updated every frame for the robust target tracking against illumination change. As for a single image, however, most of studies employ a fixed skin color model which may result in low detection rate or high false positive errors. In this paper, we propose a novel method for effective skin color segmentation in a single image, which modifies the conditions for skin color segmentation iteratively by the image feedback of segmented skin color region in a given image.

• Journal of Korea Multimedia Society
• /
• v.5 no.2
• /
• pp.176-190
• /
• 2002

According as vehicles increases, system such as Advanced Drivers Assistance System(ADAS ) to inform forward situation to driver is required. In this paper, we proposes method to detect forward vehicles and lane from sequential color images by basis process to inform forward situation to driver. We detect a front vehicle using that shadow area exists on part under vehicles and that road area occupies many parts even if road traffic is confused. We detect lane information using that lane part is white order by reverse characteristic of shadow area. This method shows good result in case road is confused or there is direction indication to road. HSV color space is selected for color modeling. This method uses saturation component and value component in HSV color model to detect vehicles and lane. It uses statistics features of HSV component and position to know whether detected vehicles area is vehicles such as vehicles previous frame. To verify the effects of the proposed method, we capture the road images with notebook and CCD camera for PC and Present the results such as processing time, accuracy and vehicles detection against the images.

• Journal of Digital Convergence
• /
• v.10 no.7
• /
• pp.123-128
• /
• 2012

In this paper, we propose a landmark-detection system of object for more accurate object recognition. The landmark-detection system of object becomes divided into a learning stage and a detection stage. A learning stage is created an interest-region model to set up a search region of each landmark as pre-information necessary for a detection stage and is created a detector by each landmark to detect a landmark in a search region. A detection stage sets up a search region of each landmark in an input image with an interest-region model created in the learning stage. The proposed system uses Fast Fourier Transform to detect landmark, because the landmark-detection is fast. In addition, the system fails to track objects less likely. After we developed the proposed method was applied to environment video. As a result, the system that you want to track objects moving at an irregular rate, even if it was found that stable tracking. The experimental results show that the proposed approach can achieve superior performance using various data sets to previously methods.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.49 no.9
• /
• pp.113-127
• /
• 2012

Face detection methods using image processing have been proposed in many different ways. Generally, the most widely used method for face detection is an Adaboost that is proposed by Viola and Jones. This method uses Haar-like feature for image learning, and the detection performance depends on the learned images. It is well performed to detect face images within a certain distance range, but if the image is far away from the camera, face images become so small that may not detect them with the pre-learned Haar-like feature of the face image. In this paper, we propose the far distance face detection method that combine the Aadaboost of Viola-Jones with a saliency map and user's attention information. Saliency Map is used to select the candidate face images in the input image, face images are finally detected among the candidated regions using the Adaboost with Haar-like feature learned in advance. And the user's eye-tracking information is used to select the interest regions. When a subject is so far away from the camera that it is difficult to detect the face image, we expand the small eye gaze spot region using linear interpolation method and reuse that as input image and can increase the face image detection performance. We confirmed the proposed model has better results than the conventional Adaboost in terms of face image detection performance and computational time.

• Journal of rehabilitation welfare engineering & assistive technology
• /
• v.10 no.2
• /
• pp.141-146
• /
• 2016

In this paper, we apply limited area mask operation and active contour model to accurately detect tongue area outline in tongue diagnosis system. To accurately analyze the properties of the tongue, first, the tongue area to be detected. Therefore an effective segmentation method for detecting the edge of tongue is very important. It experimented with tongue image DB consists of 20~30 students 30 people. Experiments on real tongue image show the good performance of this method. Experimental results show that the proposed method extracts object boundaries more accurately than existing methods without mask operation.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.2
• /
• pp.149-154
• /
• 2014

Detection performance of fire detection in the outdoor depends on weather conditions, the shadow by the movement of the sun, or illumination changes. In this paper, a smoke detection in conjunction with a robust background estimate algorithm to environment change in the outdoor in daytime is proposed. Gaussian Mixture Model (GMM) is applied as background estimation, and also, statistical characteristics of smoke is applied to detect the smoke for separated candidate region. Through the experiments with input videos obtained from a various weather conditions, the proposed algorithms were useful to detect smoke in the outdoor.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.05a
• /
• pp.706-707
• /
• 2015

최근 인터넷 사용이 급증함에 따라 통신망을 통한 악성코드의 감염 경로가 다양해지고 있다. 특히, 봇(Bot)에 의한 공격은 주로 C&C(command-and-control)서버에서 이루어지는데, C&C서버가 IP 형태로 운영되므로 IP를 차단하는 방식을 통해 보안을 유지할 수밖에 없었다. 그러나 공격자들 역시 이러한 서버 차단을 회피하기 위해 우회적인 방법으로 접속을 시도하는 등 차츰 지능화되고 있다. 이러한 악성코드는 사용자의 시스템에 침입하면, 실행이 되는 동안 일반적인 검출방법으로는 검출해 내기가 쉽지 않다. 따라서 본 논문에서는 악성코드 감염에 의한 피해 확산을 방지하기 위해 보안에 이전트 기반의 악성프로세스 검출시스템 모델을 제시하고자 한다.

• Annual Conference of KIPS
• /
• 2018.10a
• /
• pp.749-752
• /
• 2018

깊은 신경망 모델을 이용한 차량 번호판 검출과 번호판 문자 인식 시스템을 제안한다. 차량 번호판 인식 시스템은 세 가지 종류의 깊은 신경망 모델로 구성된다. 기존의 영상처리 기반의 차량 번호판 검출과 문자 인식을 전부 신경망으로 대체함으로써 영상의 밝기, 회전, 왜곡 등의 변형에 강인한 성능을 얻을 수 있다. 차량 번호판 검출률은 99.3%, 문자 영역 검출률은 99%, 문자 인식률을 98.5%를 얻었다.

• Annual Conference of KIPS
• /
• 2019.10a
• /
• pp.952-953
• /
• 2019

최근 딥러닝으로 진행되는 광학 문자 인식 분야는 대부분 단어 단위로 인식하는 것으로 글자 단위의 영역을 검출하는 데에는 적합하지 못하다. 본 연구는 각 글자의 영역을 검출하기 위해 기존의 딥러닝을 이용한 광학 문자 인식 절차인 단어 분리 과정과 단어 인식 과정을 유지하면서 어텐션 중심을 이용하여 각 글자의 영역을 보다 정확하게 검출하는 것을 목표로 한다. 제안하는 모델은 CRAFT 와 Attention Network 를 사용한 OCR 과정을 확장한 모델로 각 단어 문자열 결과물에 각 글자의 영역을 추가로 나타내게 되며 각 글자와 라벨 간의 IOU 평균은 0.671 로 나타났다.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.465-469
• /
• 2019

자연어 처리를 위해서 두 문장의 의미 유사성을 분석하는 것은 아주 중요하다. 이 논문은 패러프레이즈 검출 태스크를 수행하기 위한 Paraphrase-BERT를 제안한다. 우선 구글이 제안한 사전 학습된 BERT를 그대로 이용해서 패러프레이즈 데이터 (MRPC)를 가지고 파인 튜닝하였고 추가적으로 최근에 구글에서 새로 발표한 Whole Word Masking 기술을 사용하여 사전 학습된 BERT 모델을 새롭게 파인 튜닝하였다. 그리고 마지막으로 다중 작업 학습을 수행하여 성능을 향상시켰다. 구체적으로 질의 응답 태스크와 패러프레이즈 검출 태스크를 동시에 학습하여 후자가 더 잘 수행될 수 있도록 하였다. 결과적으로 점점 더 성능이 개선되었고 (11.11%의 정확도 향상, 7.88%의 F1 점수 향상), 향후 작업으로 파인 튜닝하는 방법에 대해서 추가적으로 연구할 계획이다.