• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.1
• /
• pp.85-92
• /
• 2020

Deep learning shows outstanding performance in image and video analysis, such as object classification, object detection and semantic segmentation. In this paper, it is analyzed that the performances of deep learning models can be affected by characteristics of train dataset. It is proposed as a method for selecting activation function and optimization algorithm of deep learning to classify facial expression. Classification performances are compared and analyzed by applying various algorithms of each component of deep learning model for CK+, MMI, and KDEF datasets. As results of simulation, it is shown that genetic algorithm can be an effective solution for optimizing components of deep learning model.

• 한국HCI학회:학술대회논문집
• /
• 2007.02a
• /
• pp.1057-1062
• /
• 2007

얼굴표정 인식은 심리학 연구, 얼굴 애니메이션 합성, 로봇공학, HCI(Human Computer Interaction) 등 다양한 분야에서 중요성이 증가하고 있다. 얼굴표정은 사람의 감정 표현, 관심의 정도와 같은 사회적 상호작용에 있어서 중요한 정보를 제공한다. 얼굴표정 인식은 크게 정지영상을 이용한 방법과 동영상을 이용한 방법으로 나눌 수 있다. 정지영상을 이용할 경우에는 처리량이 적어 속도가 빠르다는 장점이 있지만 얼굴의 변화가 클 경우 매칭, 정합에 의한 인식이 어렵다는 단점이 있다. 동영상을 이용한 얼굴표정 인식 방법은 신경망, Optical Flow, HMM(Hidden Markov Models) 등의 방법을 이용하여 사용자의 표정 변화를 연속적으로 처리할 수 있어 실시간으로 컴퓨터와의 상호작용에 유용하다. 그러나 정지영상에 비해 처리량이 많고 학습이나 데이터베이스 구축을 위한 많은 데이터가 필요하다는 단점이 있다. 본 논문에서 제안하는 실시간 얼굴표정 인식 시스템은 얼굴영역 검출, 얼굴 특징 검출, 얼굴표정 분류, 아바타 제어의 네 가지 과정으로 구성된다. 웹캠을 통하여 입력된 얼굴영상에 대하여 정확한 얼굴영역을 검출하기 위하여 히스토그램 평활화와 참조 화이트(Reference White) 기법을 적용, HT 컬러모델과 PCA(Principle Component Analysis) 변환을 이용하여 얼굴영역을 검출한다. 검출된 얼굴영역에서 얼굴의 기하학적 정보를 이용하여 얼굴의 특징요소의 후보영역을 결정하고 각 특징점들에 대한 템플릿 매칭과 에지를 검출하여 얼굴표정 인식에 필요한 특징을 추출한다. 각각의 검출된 특징점들에 대하여 Optical Flow알고리즘을 적용한 움직임 정보로부터 특징 벡터를 획득한다. 이렇게 획득한 특징 벡터를 SVM(Support Vector Machine)을 이용하여 얼굴표정을 분류하였으며 추출된 얼굴의 특징에 의하여 인식된 얼굴표정을 아바타로 표현하였다.

• Korean Journal of Cleft Lip And Palate
• /
• v.13 no.2
• /
• pp.85-92
• /
• 2010

Vascular malformations (VMs) in the head and neck region are present at birth and grow commensurately with the child, they can result in significant cosmetic problems for the patient, and some may lead to even serious life threatening hemorrhage. Although the molecular mechanisms underlying the formation of these VMs remain unclear, lesions are known to result from abnormal development and morphogenesis. Histologically, there are no evidence of cellular proliferation, but rather progressive dilatation of abnormal channels, which VMs are designated to their prominent channel types such as capillary, venous, lymphatic, arterial, and combined malformations. VMs with an arterial component are rheologically fast-flow, whereas capillary, lymphatic, and venous components are slow-flow. In this article, we review the clinical presentations, diagnosis, and management of VMs of facial regions with author's embolization and surgical treatment cases.

• Journal of Korea Multimedia Society
• /
• v.19 no.8
• /
• pp.1310-1319
• /
• 2016

Face detection is the first step in a wide range of face applications. However, detecting faces in the wild is still a challenging task due to the wide range of variations in pose, scale, and occlusions. Recently, many deep learning methods have been proposed for face detection. However, further improvements are required in the wild. Another important issue to be considered in the face detection is the computational complexity. Current state-of-the-art deep learning methods require a large number of patches to deal with varying scales and the arbitrary image sizes, which result in an increased computational complexity. To reduce the complexity while achieving better detection accuracy, we propose a fully convolutional network-based face detection that can take arbitrarily-sized input and produce feature maps (heat maps) corresponding to the input image size. To deal with the various face scales, a multi-scale network architecture that utilizes the facial components when learning the feature maps is proposed. On top of it, we design multi-task learning technique to improve detection performance. Extensive experiments have been conducted on the FDDB dataset. The experimental results show that the proposed method outperforms state-of-the-art methods with the accuracy of 82.33% at 517 false alarms, while improving computational efficiency significantly.

• Archives of Plastic Surgery
• /
• v.41 no.2
• /
• pp.163-170
• /
• 2014

Background The septal cartilage is the most useful donor site for autologous cartilage graft material in rhinoplasty. For successful nasal surgery, it is necessary to understand the developmental process of the nasal septum and to predict the amount of harvestable septal cartilage before surgery. Methods One hundred twenty-three Korean patients who underwent three-dimensional (3D) facial bone computed tomography (CT) were selected for evaluation of the midsagittal view of the nasal septum. Multiple parameters such as the area of each component of the nasal septum and the amount of harvestable septal cartilage were measured using Digimizer software. Results The area of the total nasal septum showed rapid growth until the teenage years, but thereafter no significant change throughout the lifetime. However, the development of the septal cartilage showed a gradual decline due to ossification changes with aging after puberty in spite of a lack of change in the total septal area. The area of harvestable septal cartilage in young adults was $549.84{\pm}151.26mm^2$ and decreased thereafter with age. Conclusions A 3D facial bone CT scan can provide valuable information on the septal cartilage graft before rhinoplasty. Considering the developmental process of the septal cartilage identified in this study, septal surgery should not be performed until puberty due to the risk of nasal growth impairment. Furthermore, in elderly patients who show a decreased cartilage area due to ossification changes, septal cartilage harvesting should be performed carefully due to the risk of saddle nose deformity.

• Archives of Plastic Surgery
• /
• v.38 no.6
• /
• pp.903-906
• /
• 2011

Purpose: Accessory tragus is a fairly common congenital malformation and usually located at pretragal area. Surgical removal is a common treatment of accessory tragus irrespective of location and morphology. Most accessory tragi do not have depression site around them, but some do. So in those cases, simple surgical excision was not enough to promote the aesthetic facial appearance. For depression site remodeling, the excess amount of skin and cartilage need to be remained partially instead of total excision. This method can achieve the symmetric contour of pretragal area. The authors excised the epidermis and cartilaginous tissue totally and remained the dermis for reconstruction of the depression site around accessory tragus. The depression site is filled with dermal turnover flap. The purpose of this report is to present new idea to promote cosmetic result in treatment of accessory tragus containing the depression site. Methods: Two patients had a pair of accessory tragi at pretragal area. One was a common featured accessory tragus, but the other was different. Depression site was found around accessory tragus. After epidermis and cartilaginous tissue were removed from it, dermis component was used as turnover flap for reconstruction of depression site. Results: After accessory tragus was removed and depression site was reconstructed, facial contour and cosmetic result was achieved. Complication such as flap necrosis and wound dehiscence was not observed. Conclusion: The accessory tragus has variant morphology and degree of invasive depth. And some has a depression site around them. In those cases, simple surgical removal results in morphological distorsion and do not promote facial symmetry. The authors suggest dermal turnover flap as reconstruction method of the depression site. This method improves both surgical outcome and cosmetic result.

• Archives of Plastic Surgery
• /
• v.36 no.2
• /
• pp.205-210
• /
• 2009

Purpose: Corrugator supercilii muscle pulls eyebrow to inferomedial direction and produces the vertical component of the glabellar line formation. Current techniques for eliminating of glabellar frown include direct resection of corrugators and botulinum toxin injection. Muscle resection in endoscopic face lift procedure is relatively complex and has many disadvantages ranging from possible nerve injury, postoperative edema, pain and a long recovery period. The Botox treatment on the other hand is much more simple in technique but has a short duration of action. The authors have attempted new ways of finding improved treatment of the glabellar frown by selectively blocking of motor nerves innervating the corrugator supercili muscle by using radiofrequency ablation technique. Methods: A total of 80 patients were recruited in our study during the period between Feb. 2007 to June 2008. A probe was introduced from the supraorbital ridge and advanced to the corrugator supercilii muscle. Nerve stimulator was then used to locate the nerve innervating the corrugator and radiofrequency ablation of the nerve was done. Results: In all patients, there were marked improvement in glabellar frown after treatment. There were no reported cases of any relapses during the follow up period. No complication was noted such as facial nerve injury. No patient complained of any adverse symptoms other than slight discomfort due to swelling of the operation site. Conclusion: The treatment of glabellar frown lines using selective nerve block with radiofrequency ablation was not only less invasive but also excellent in surgical outcomes.

• The Journal of the Korea Contents Association
• /
• v.10 no.1
• /
• pp.19-26
• /
• 2010

This paper proposes a classifier based on rectangular feature to detect face in real time. The goal is to realize a strong detection algorithm which satisfies both efficiency in calculation and detection performance. The proposed algorithm consists of the following three stages: Feature creation, classifier study and real time facial domain detection. Feature creation organizes a feature set with the proposed five rectangular features and calculates the feature values efficiently by using SAT (Summed-Area Tables). Classifier learning creates classifiers hierarchically by using the AdaBoost algorithm. In addition, it gets excellent detection performance by applying important face patterns repeatedly at the next level. Real time facial domain detection finds facial domains rapidly and efficiently through the classifier based on the rectangular feature that was created. Also, the recognition rate was improved by using the domain which detected a face domain as the input image and by using PCA and KNN algorithms and a Class to Class rather than the existing Point to Point technique.

• The KIPS Transactions:PartB
• /
• v.9B no.6
• /
• pp.853-862
• /
• 2002

Face Detection can be defined as follows : Given a digitalized arbitrary or image sequence, the goal of face detection is to determine whether or not there is any human face in the image, and if present, return its location, direction, size, and so on. This technique is based on many applications such face recognition facial expression, head gesture and so on, and is one of important qualify factors. But face in an given image is considerably difficult because facial expression, pose, facial size, light conditions and so on change the overall appearance of faces, thereby making it difficult to detect them rapidly and exactly. Therefore, this paper proposes fast and exact face detection which overcomes some restrictions by using neural network. The proposed system can be face detection irrelevant to facial expression, background and pose rapidily. For this. face detection is performed by neural network and detection response time is shortened by reducing search region and decreasing calculation time of neural network. Reduced search region is accomplished by using skin color segment and frame difference. And neural network calculation time is decreased by reducing input vector sire of neural network. Principle Component Analysis (PCA) can reduce the dimension of data. Also, pose estimates in extracted facial image and eye region is located. This result enables to us more informations about face. The experiment measured success rate and process time using the Squared Mahalanobis distance. Both of still images and sequence images was experimented and in case of skin color segment, the result shows different success rate whether or not camera setting. Pose estimation experiments was carried out under same conditions and existence or nonexistence glasses shows different result in eye region detection. The experiment results show satisfactory detection rate and process time for real time system.

• The KIPS Transactions:PartB
• /
• v.16B no.4
• /
• pp.299-308
• /
• 2009

AAM(Active Appearance Model) is an algorithm to extract face feature points with statistical models of shape and texture information based on PCA(Principal Component Analysis). This method is widely used for face recognition, face modeling and expression recognition. However, the detection performance of AAM algorithm is sensitive to initial value and the AAM method has the problem that detection error is increased when an input image is quite different from training data. Especially, the algorithm shows high accuracy in case of closed lips but the detection error is increased in case of opened lips and deformed lips according to the facial expression of user. To solve these problems, we propose the improved AAM algorithm using lip feature points which is extracted based on a new lip detection algorithm. In this paper, we select a searching region based on the face feature points which are detected by AAM algorithm. And lip corner points are extracted by using Canny edge detection and histogram projection method in the selected searching region. Then, lip region is accurately detected by combining color and edge information of lip in the searching region which is adjusted based on the position of the detected lip corners. Based on that, the accuracy and processing speed of lip detection are improved. Experimental results showed that the RMS(Root Mean Square) error of the proposed method was reduced as much as 4.21 pixels compared to that only using AAM algorithm.