• Journal of the Korea Society of Computer and Information
• /
• v.26 no.11
• /
• pp.85-92
• /
• 2021

In this paper, a method for effectively detecting rotated face and rotation angle regardless of the rotation angle is proposed. Rotated face detection is a challenging task, due to the large variation in facial appearance. In the proposed polar coordinate transformation, the spatial information of the facial components is maintained regardless of the rotation angle, so there is no variation in facial appearance due to rotation. Accordingly, features such as HOG, which are used for frontal face detection without rotation but have rotation-sensitive characteristics, can be effectively used in detecting rotated face. Only the training data in the frontal face is needed. The HOG feature obtained from the polar coordinate transformed images is learned using SVM and rotated faces are detected. Experiments on 3600 rotated face images show a rotation angle detection rate of 97.94%. Furthermore, the positions and rotation angles of the rotated faces are accurately detected from images with a background including multiple rotated faces.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.7
• /
• pp.896-902
• /
• 2021

Rotated face detection is required in many applications but still remains as a challenging task, due to the large variations of face appearances. In this paper, a polar coordinate transform that is not affected by rotation is proposed. In addition, a method for effectively detecting rotated faces using the transformed image has been proposed. The proposed polar coordinate transform maintains spatial information between facial components such as eyes, mouth, etc., since the positions of facial components are always maintained regardless of rotation angle, thereby eliminating rotation effects. Polar coordinate transformed images are trained using AdaBoost, which is used for frontal face detection, and rotated faces are detected. We validate the detected faces using LBP that trained the non-face images. Experiments on 3600 face images obtained by rotating images in the BioID database show a rotating face detection rate of 96.17%. Furthermore, we accurately detected rotated faces in images with a background containing multiple rotated faces.

• Journal of KIISE:Software and Applications
• /
• v.34 no.10
• /
• pp.945-952
• /
• 2007

The face detection is a necessary first-step in the face recognition systems, with the purpose of localizing and extracting face regions from input images. But it is not a simple problem, because faces have many variations such as scale, rotation and lighting condition. In this paper, we propose a novel method to detect not only frontal faces but also partial rotated faces in still images. Firstly, we produce the eye candidates in the sub-regions of an input image to detect rotated faces. Secondly, the eye candidates are used to measure the angles of rotated faces. Thirdly, we are able to derotate the rotated face then put it to Bayesian classifier. We make an experiment with rotated multiracial face and show the good results in this paper.

• Journal of Korea Multimedia Society
• /
• v.9 no.2
• /
• pp.234-243
• /
• 2006

The recent rapid development of HCI and surveillance technologies has brought great interests in application systems to process faces. Much of research efforts in these systems has been primarily focused on such areas as face recognition, facial expression analysis and facial feature extraction. However, not many approaches have been reported toward face direction detection. This paper proposes a method to detect the direction of a face using a facial feature called facial triangle, which is formed by two eyebrows and the lower lip. Specifically, based on the single monocular view of the face, the proposed method introduces very simple formulas to estimate the horizontal or vertical rotation angle of the face. The horizontal rotation angle can be calculated by using a ratio between the areas of left and right facial triangles, while the vertical angle can be obtained from a ratio between the base and height of facial triangle. Experimental results showed that our method makes it possible to obtain the horizontal angle within an error tolerance of ${\pm}1.68^{\circ}$, and that it performs better as the magnitude of the vertical rotation angle increases.

• Journal of the Korea Society of Computer and Information
• /
• v.16 no.8
• /
• pp.49-56
• /
• 2011

Recently, animation contents has become extensively available along with the development of cultural industry. In this paper, we propose a method to analyze a face of animation character and extract 3D rotational information of the face. The suggested method first generates a dominant color model of a face by learning the face image of animation character. Our system then detects the face and its components with the model, and establishes two coordinate systems: base coordinate system and target coordinate system. Our system estimates three dimensional rotational information of the animation character face using the geometric relationship of the two coordinate systems. Finally, in order to visually represent the extracted 3D information, a 3D face model in which the rotation information is reflected is displayed. In experiments, we show that our method can extract 3D rotation information of a character face reasonably.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2010.07a
• /
• pp.195-197
• /
• 2010

본 논문은 기존 평행좌표를 이용하는 얼굴 영상 대신 극좌표계 변환을 이용한 얼굴 영상을 이용하여 회전에 강인한 얼굴인식 방법을 제안한다. 극좌표계 변환 방법은 얼굴의 중심부분의 한 점을 극으로 삼아 이 점을 기준으로 360도 각 방향으로 일정 길이만큼 얼굴 영상을 샘플링 하여 새로운 얼굴 영상을 제작하는 방법이다. 이 극좌표계 변환 방법을 이용해 재구성된 영상에 대해 회귀( regression )문제 해결을 위해 변형된 LDA인 LDAr(LDA for regression)을 이용하여 얼굴의 중심부분의한 점인 극을 중심으로 임의의 각도로 회전된 영상의 회전 정도를 추정하여 이를 정규화 시키는 방법을 통해 얼굴 인식의 인식률을 향상시키고자 한다. LDAr은 LDA의 기본개념인 각 클래스 간 떨어진 정도를 최대화하는 것이 목적으로 클래스간 분산과 클래스내 분산의 비율을 최대화 하는 방법을 응용하여 이를 회귀문제에 적용할 수 있게 변형을 한 것이다. 즉, LDAr은 목표값(target)의 차이가 큰 샘플들과 목표값의 차이가 작은 샘플들 간의 거리의 비율을 최대화 하는 것을 목적으로 하게 된다. 제안된 방법을 Yale데이터에 적용하여 임의의 각도로 회전시킨 영상에 대해 회전 각도를 정확히 찾아내는 것을 확인할 수 있었다.

• Journal of the Korea Society of Computer and Information
• /
• v.16 no.1
• /
• pp.53-59
• /
• 2011

In many face recognition applications such as security systems, it is assumed that upright faces are given to the system. In order for the system to be used in more general environments, the system should be able to deal with the rotated faces properly. It is a generally used approach to rotate the face detection window and apply face detector repeatedly to detect a rotated face in the given image. But such an approach requires a lot of computation time. In this paper, a method of extracting the axis of symmetry for a given set of points is proposed. The axis of symmetry for the edge points in the face detection window is extracted in a way that is fast and accurate, and the face detector is applied only for that direction. It is shown that the mean and standard deviation of the symmetry detection error is $0^{\circ}$ and $3^{\circ}$ respectively, for the database used.

• The KIPS Transactions:PartB
• /
• v.10B no.4
• /
• pp.465-472
• /
• 2003

For an effective pre-treatment process of a face input image, it is necessary to detect each of face components, calculate the face area, and estimate the rotary angle of the face. A proposed method of this study can estimate an robust result under such renditions as some different levels of illumination, variable fate sizes, fate rotation angels, and background color similar to skin color of the face. The first step of the proposed method detects the estimated face area that can be calculated by both adapted skin color Information of the band-wide HSV color coordinate converted from RGB coordinate, and skin color Information using histogram. Using the results of the former processes, we can detect a lip area within an estimated face area. After estimating a rotary angle slope of the lip area along the X axis, the method determines the face shape based on face information. After detecting eyes in face area by matching a partial template which is made with both eyes, we can estimate Y axis rotary angle by calculating the eye´s locations in three dimensional space in the reference of the face area. As a result of the experiment on various face images, the effectuality of proposed algorithm was verified.

• The KIPS Transactions:PartB
• /
• v.8B no.1
• /
• pp.28-34
• /
• 2001

컴퓨터 시각 인식 방법을 이용하여 얼굴의 3차원 움직임 량을 추정하고자 하는 연구는 가상 현실 환경에서 얼굴 움직임에 의한 3차원 그래픽 화면 조정, 시뮬레이터에서의 훈련자 얼굴 움직임에 의한 화면 조정 및 모니터상의 시선 위치 파악 등을 위해 필수적으로 요구되는 기술로서 최근 활발히 연구되고 있다. 기존에 얼굴의 3차원 움직임 량을 추정하고자 하는 연구들은 대부분 확장 칼만 필터(extended kalman filter)를 이용하였으나, 이러한 방법은 필터의 초기 값을 정확하게 설정해야하는 제약 요소를 갖고 있으며, 또한 얼굴의 회전 방향 변화 시 이에 대처하지 못하는 경우 역시 종종 발생한다. 본 논문에서는 이러한 문제점을 해결하기 위하여 확장 칼만 필터의 변형 형태인 반복적 확장 칼만 필터를 이용하여 얼굴의 3차원 움직임 량을 추정하였다. 반복적 확장 칼만 필터에서는 확장 칼만 필터에서 계산되어 나오는 추정 오차 공분산 행렬 값이 미리 정해진 임계치보다 커지는 경우, 현재 얼굴의 움직임 량을 제대로 추정하지 못하고 있는 것으로 판단하여 칼만 필터에서 사용하는 회전 및 이동 속도, 그리고 회전 및 이동 각 속도를 변형함으로써 얼굴의 움직임 량을 정확하게 추정할 수 있도록 하는 방법이다. 실험 결과 반복적 확장 칼만 필터를 사용하였을 경우에 얼굴의 급격한 회전 방향 변화에도 얼굴의 3차원 움직임 량을 정확하게 추정할 수 있음을 알 수 있었다.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.36S no.8
• /
• pp.88-100
• /
• 1999

Detecting human face and facial landmarks automatically in an image is as essential step to a fully automatic face recognition system. In this paper, we present a new approach to detect automatically face and its eyes of input image with scale and rotation variations of faces by using an intensity based template matching with a single log-polar face template. In a template-based matching it is necessary to normalize the scale changes and rotations of an input image to a template ones. The log-polar mapping which simulates space-variant human visual system converts scale changes and rotations of input image into constant horizontal and cyclic vertical shifts in the output plane. Intelligent use of this property allows us to shift of the candidate log-polar faces mapped at various fixation points of an input image to be matched to a template over the log-polar plane. Thus, the proposed method eliminates the need of adapting multitemplate and multiresolution schemes, which inevitably give rise to intensive computation involved to cope with scale and rotation variations of faces. Through this scale and rotation involved to cope with scale and method can lead to detecting face and its eyes simultaneously. Experimental results on a database of 795 images show over 98% detection rate.