• 한국정보과학회논문지:소프트웨어및응용
• /
• 제34권1호
• /
• pp.52-58
• /
• 2007

본 논문에서는 실내 환경에서 동적 스테레오 카메라(active stereo camera)를 이용한 새로운 인체 실루엣 추출 방법을 제안한다. 제안한 알고리즘의 주된 응용분야는 이동 로봇 플랫폼에서의 인체 실루엣을 이용한 제스처 인식이다. 먼 거리에서 움직이는 객체를 분할(segmentation)하는 데에는 저해상도, 그림자, 스테레오 정합의 불확실성, 배경과 객체의 색 분포의 불안정성 등과 같은 다양한 문제를 내포한다. 우리는 먼저 이미지 분할 기법과 스테레오 정보를 이용하여 신뢰도 높은 객체와 배경 영역을 추정하였다. 이렇게 추정된 영역을 적절히 그래프 컷(graph cut)에 활용하는 방식을 고안함으로써 주변 환경의 변화에 강인한 인체 실루엣 추출을 가능하게 하였다. 제안한 방식은 실내에서 펜-틸트(pan-tilt) 스테레오 카메라로 획득된 비디오 데이타를 대상으로 실험하였으며, 색, 색과 스테레오, 색과 대비 정보를 기반으로 한 방법들과 비교 실험한 결과 정확도가 많이 향상된 것을 확인할 수 있었다.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2001년도 제14회 신호처리 합동 학술대회 논문집
• /
• pp.139-142
• /
• 2001

This paper presents a highly fast and accurate facial region extraction method by using the skin-color-reference map and motion information. First, we construct the robust skin-color-reference map and eliminate the background in image by this map. Additionally, we use the motion information for accurate and fast detection of facial region in image sequences. Then we further apply region growing in the remaining areas with the aid of proposed criteria. The simulation results show the improvement in execution time and accurate detection.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2007년도 춘계학술대회 학술발표 논문집 제17권 제1호
• /
• pp.444-447
• /
• 2007

본 논문에서는 비디오에서 자막을 자동 추출하기 위한 강건한 자막 분리 알고리즘을 제안한다. 주어진 비디오에서 자막이 존재할 가능성이 있는 프레임에 대해 자막 후보 영역의 위치를 찾은 다음, 자막 후보 영역으로부터 강건하게 자막을 추출한다. 추출된 자막 후보 영역에 대해 Dampoint labeling을 수행하여 자막과 비슷한 색상을 갖는 배경을 제거하고, 마지막으로 기하학적 검증을 통해 최종적으로 자막 여부를 판별한다. 제안된 방법을 여러 장르의 비디오에 대해 적용 결과 복잡한 배경을 갖는 비디오에서 자막을 강건하게 추출함을 실험을 통해 확인하였다.

• IEIE Transactions on Smart Processing and Computing
• /
• 제5권3호
• /
• pp.153-163
• /
• 2016

Local feature extraction methods for images and videos are widely applied in the fields of image understanding and computer vision. However, robust features are detected differently when using the latest feature detectors and descriptors because of diverse image environments. This paper analyzes various feature extraction methods by summarizing algorithms, specifying properties, and comparing performance. We analyze eight feature extraction methods. The performance of feature extraction in various image environments is compared and evaluated. As a result, the feature detectors and descriptors can be used adaptively for image sequences captured under various image environments. Also, the evaluation of feature detectors and descriptors can be applied to driving assistance systems, closed circuit televisions (CCTVs), robot vision, etc.

• 제어로봇시스템학회논문지
• /
• 제17권4호
• /
• pp.346-352
• /
• 2011

This paper presents a system of road extraction for traffic images from a single camera. The road in the images is subject to large changes in appearance because of environmental effects. The proposed system is based on the integration of color correlograms and random forest. The color correlogram depicts the color properties of an image properly. Using the random forest, road extraction is formulated as a learning paradigm. The combined effects of color correlograms and random forest create a robust system capable of extracting the road in very changeable situations.

• International Journal of Contents
• /
• 제2권3호
• /
• pp.27-32
• /
• 2006

We present an effective algorithm for automatic tracing of retinal vessel structure and vascular landmark extraction of bifurcations and ending points. In this paper we deal with vascular patterns from RGB images for personal identification. Vessel tracing algorithms are of interest in a variety of biometric and medical application such as personal identification, biometrics, and ophthalmic disorders like vessel change detection. However eye surface vasculature tracing in RGB images has many problems which are subject to improper illumination, glare, fade-out, shadow and artifacts arising from reflection, refraction, and dispersion. The proposed algorithm on vascular tracing employs multi-stage processing of ten-layers as followings: Image Acquisition, Image Enhancement by gray scale retinal image enhancement, reducing background artifact and illuminations and removing interlacing minute characteristics of vessels, Vascular Structure Extraction by connecting broken vessels, extracting vascular structure using eight directional information, and extracting retinal vascular structure, and Vascular Landmark Extraction by extracting bifurcations and ending points. The results of automatic retinal vessel extraction using jive different thresholds applied 34 eye images are presented. The results of vasculature tracing algorithm shows that the suggested algorithm can obtain not only robust and accurate vessel tracing but also vascular landmarks according to thresholds.

• 전자공학회논문지SC
• /
• 제45권2호
• /
• pp.8-17
• /
• 2008

본 논문에서는 매우 다양한 조건하에서 취득된 영상으로부터 차량 번호판 영역을 추출하는 방법을 제안한다. 제안하는 방법은 색상정보와 에지정보를 보완적으로 사용함으로써 다양한 조명조건은 물론 압축영상에서 흔히 나타나는 영상의 블록화 현상에 강건하게 번호판을 추출한다. 번호판 기울기 각도 보정은 탐색된 영역만을 허프 변환을 통해 각도를 추정, 보정함으로써 연산량을 줄였다. 최종 번호판 후보지 결정은 대상 영역의 가로 대 세로 비율 평가, 수평 스캔하여 영점 교차되는 양상 평가, 레이블링 방법을 이용한 연결 성분 수 측정 등의 방법을 이용하여 결정한다. 다양한 시간대에 수집된 차량 영상을 대상으로 수행된 실험을 통해 제안하는 방법이 강건하게 번호판을 추출 할 수 있음을 확인하였다. 실제 주정차 단속 영상을 대상으로 실험한 결과 94.9%의 번호판 추출 결과를 얻었다.

• ETRI Journal
• /
• 제29권2호
• /
• pp.179-188
• /
• 2007

We propose a multi-strategic concept-spotting approach for robust spoken language understanding of conversational Korean in a hostile recognition environment such as in-car navigation and telebanking services. Our concept-spotting method adopts a partial semantic understanding strategy within a given specific domain since the method tries to directly extract predefined meaning representation slot values from spoken language inputs. In spite of partial understanding, we can efficiently acquire the necessary information to compose interesting applications because the meaning representation slots are properly designed for specific domain-oriented understanding tasks. We also propose a multi-strategic method based on this concept-spotting approach such as a voting method. We present experiments conducted to verify the feasibility of these methods using a variety of spoken Korean data.

• 음성과학
• /
• 제11권1호
• /
• pp.7-20
• /
• 2004

In this paper, we propose a two-step noise compensation algorithm in feature extraction for achieving robust speech recognition. The proposed method frees us from requiring a priori information on noisy environments and is simple to implement. First, in frequency domain, the Harmonics-based Spectral Subtraction (HSS) is applied so that it reduces the additive background noise and makes the shape of harmonics in speech spectrum more pronounced. We then apply a judiciously weighted variance Feature Vector Normalization (FVN) to compensate for both the channel distortion and additive noise. The weighted variance FVN compensates for the variance mismatch in both the speech and the non-speech regions respectively. Representative performance evaluation using Aurora 2 database shows that the proposed method yields 27.18% relative improvement in accuracy under a multi-noise training task and 57.94% relative improvement under a clean training task.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제11권1호
• /
• pp.345-359
• /
• 2017

Optical Character Recognition (OCR) that has been a main research topic of computer vision and artificial intelligence now extend its applications to detection of text area from video or image contents taken by camera devices and retrieval of text information from the area. This paper aims to implement a binarization algorithm that removes user intervention and provides robust performance to outdoor lights by using TopHat algorithm and channel transformation technique. In this study, we particularly concentrate on text information of outdoor signboards and validate our proposed technique using those data.