• Science of Emotion and Sensibility
• /
• v.16 no.1
• /
• pp.95-106
• /
• 2013

The study compares accuracies between humans and computer algorithms in the discrimination of spontaneous smiles from posed smiles. For this purpose, subjects performed two tasks, one was judgment with single pictures and the other was judgment with pair comparison. At the task of judgment with single pictures, in which pictures of smiling facial expression were presented one by one, subjects were required to judge whether smiles in the pictures were spontaneous or posed. In the task for judgment with pair comparison, in which two kinds of smiles from one person were presented simultaneously, subjects were to select spontaneous smile. To calculate the discrimination algorithm accuracy, 8 kinds of facial features were used. To calculate the discriminant function, stepwise linear discriminant analysis (SLDA) was performed by using approximately 50 % of pictures, and the rest of pictures were classified by using the calculated discriminant function. In the task of single pictures, the accuracy rate of SLDA was higher than that of humans. In the analysis of accuracy on pair comparison, the accuracy rate of SLDA was also higher than that of humans. Among the 20 subjects, none of them showed the above accuracy rate of SLDA. The facial feature contributed to SLDA effectively was angle of inner eye corner, which was the degree of the openness of the eyes. According to Ekman's FACS system, this feature corresponds to AU 6. The reason why the humans had low accuracy while classifying two kinds of smiles, it appears that they didn't use the information coming from the eyes enough.

• Proceedings of the Korean Society for Emotion and Sensibility Conference
• /
• 2011.11a
• /
• pp.104-105
• /
• 2011

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.35-37
• /
• 2019

최근 딥 러닝의 급격한 발전과 함께 얼굴표정인식 기술이 상당한 진보를 이루었다. 그러나 기존 얼굴표정인식 기법들은 제한된 환경에서 취득한 인위적인 동영상에 대해 주로 개발되었기 때문에 실제 wild 한 환경에서 취득한 동영상에 대해 강인하게 동작하지 않을 수 있다. 이런 문제를 해결하기 위해 3D CNN, 2D CNN 그리고 RNN 의 새로운 결합으로 이루어진 Deep neural network 구조를 제안한다. 제안 네트워크는 주어진 동영상으로부터 두 가지 서로 다른 CNN 을 통해서 영상 내 공간적 정보뿐만 아니라 시간적 정보를 담고 있는 특징 벡터를 추출할 수 있다. 그 다음, RNN 이 시간 도메인 학습을 수행할 뿐만 아니라 상기 네트워크들에서 추출된 특징 벡터들을 융합한다. 상기 기술들이 유기적으로 연동하는 제안된 네트워크는 대표적인 wild 한 공인 데이터세트인 AFEW 로 실험한 결과 49.6%의 정확도로 종래 기법 대비 향상된 성능을 보인다.

• Journal of the Korea Concrete Institute
• /
• v.22 no.1
• /
• pp.113-122
• /
• 2010

Corrosion of prestressing tendons and wire fractures in grouted post-tensioned prestressed concrete bridges have been considered as a serious safety problem. In bridge evaluation the condition of prestressing tendons should be inspected, and if corroded tendons are found, the loss of tendon area should be included when we calculate the ultimate strength. In the previous study, it was evaluated that continuous acoustic monitoring techniques could be considered as a reliable non-destructive method for detecting wire fractures of fully grouted post-tensioned prestressing tendons. In the present study, an experimental test was performed for detecting wire fractures of post-tensioned prestressing tendons which are prestressed lower than current design level. A 10 m prestressed concrete beam was fabricated, which included two tendons prestressed 66 percentage and 40 percentage of tensile strength, respectively. The corrosion of two tendons was induced by an accelerated corrosion equipment and the test beam was monitored by using seven acoustic sensors and a continuous acoustic monitoring system. From each prestressing tendon, two acoustic signals of wire fractures were successfully detected and source locations were estimated within 20 mm error. Based on the test results, it is considered that continuous acoustic monitoring techniques can be applied to detect low-prestressed wire fracture in fully grouted post-tensioned prestressed concrete beams.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.49 no.4
• /
• pp.77-84
• /
• 2012

Face classification has wide applications in intelligent video surveillance, content retrieval, robot vision, and human-machine interface. Pose and expression changes, and arbitrary illumination are typical problems for face recognition. When the face is captured at a distance, the image quality is often degraded by blurring and noise corruption. This paper investigates the efficacy of multi-classifier decision level fusion for face classification based on the photon-counting linear discriminant analysis with two different cost functions: Euclidean distance and negative normalized correlation. Decision level fusion comprises three stages: cost normalization, cost validation, and fusion rules. First, the costs are normalized into the uniform range and then, candidate costs are selected during validation. Three fusion rules are employed: minimum, average, and majority-voting rules. In the experiments, unfocusing and motion blurs are rendered to simulate the effects of the long distance environments. It will be shown that the decision-level fusion scheme provides better results than the single classifier.