• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.44 no.3
• /
• pp.98-107
• /
• 2007

For high-dimensional pattern recognition, such as face classification, the small number of training samples leads to the Small Sample Size problem when the number of pattern samples is smaller than the number of dimensionality. Recently, various LDA-extensions have been developed, including LDA, PCA+LDA, and Direct-LDA, to address the problem. This paper proposes a method of improving the classification efficiency by increasing the number of (sub)-classes through pre-clustering a training set prior to the execution of Direct-LDA. In LDA (or Direct-LDA), since the number of classes of the training set puts a limit to the dimensionality to be reduced, it is increased to the number of sub-classes that is obtained through clustering so that the classification performance of LDA-extensions can be improved. In other words, the eigen space of the training set consists of the range space and the null space, and the dimensionality of the range space increases as the number of classes increases. Therefore, when constructing the transformation matrix, through minimizing the null space, the loss of discriminatve information resulted from this space can be minimized. Experimental results for the artificial data of X-OR samples as well as the bench mark face databases of AT&T and Yale demonstrate that the classification efficiency of the proposed method could be improved.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.16 no.3
• /
• pp.115-122
• /
• 2016

This paper proposes a system to detect driver's cognitive state by internal and external information of vehicle. The proposed system can measure driver's eye gaze. This is done by concept of information delivery and mutual information measure. For this study, we set up two web-cameras at vehicles to obtain visual information of the driver and front of the vehicle. We propose Gestalt principle based selective attention model to define information quantity of road scene. The saliency map based on gestalt principle is prominently represented by stimulus such as traffic signals. The proposed system assumes driver's cognitive resource allocation on the front scene by gaze analysis and head pose direction information. Then we use several feature algorithms for detecting driver's characteristics in real time. Modified census transform (MCT) based Adaboost is used to detect driver's face and its component whereas POSIT algorithms are used for eye detection and 3D head pose estimation. Experimental results show that the proposed system works well in real environment and confirm its usability.

• Journal of KIISE:Software and Applications
• /
• v.35 no.10
• /
• pp.600-607
• /
• 2008

This paper proposes a robust human detection method irrespective of their pose variation using the multiple oriented 2D elliptical filters (MO2DEFs). The MO2DEFs can detect the humans regardless of their poses unlike existing object oriented scale adaptive filter (OOSAF). To overcome OOSAF's limitation, we introduce the MO2DEFs whose shapes look like the oriented ellipses. We perform human detection by applying four different 2D elliptical filters with specific orientations to the 2D spatial-depth histogram and then by taking the thresholds over the filtered histograms. In addition, we determine the human pose by using convolution results which are computed by using the MO2DEFs. We verify the human candidates by either detecting the face or matching head-shoulder shapes over the estimated rotation. The experimental results showed that the accuracy of pose angle estimation was about 88%, the human detection using the MO2DEFs outperformed that of using the OOSAF by $15{\sim}20%$ especially in case of the posed human.

• Journal of Broadcast Engineering
• /
• v.23 no.3
• /
• pp.351-360
• /
• 2018

Human emotion recognition is a research topic that is receiving continuous attention in computer vision and artificial intelligence domains. This paper proposes a method for classifying human emotions through multiple neural networks based on multi-modal signals which consist of image, landmark, and audio in a wild environment. The proposed method has the following features. First, the learning performance of the image-based network is greatly improved by employing both multi-task learning and semi-supervised learning using the spatio-temporal characteristic of videos. Second, a model for converting 1-dimensional (1D) landmark information of face into two-dimensional (2D) images, is newly proposed, and a CNN-LSTM network based on the model is proposed for better emotion recognition. Third, based on an observation that audio signals are often very effective for specific emotions, we propose an audio deep learning mechanism robust to the specific emotions. Finally, so-called emotion adaptive fusion is applied to enable synergy of multiple networks. The proposed network improves emotion classification performance by appropriately integrating existing supervised learning and semi-supervised learning networks. In the fifth attempt on the given test set in the EmotiW2017 challenge, the proposed method achieved a classification accuracy of 57.12%.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.6
• /
• pp.43-57
• /
• 2009

Integrating 3D data acquired in multiple views is one of the most important techniques in 3D modeling. However, the existing integration methods are sensitive to registration errors and surface scanning noise. In this paper, we propose a integration algorithm using the local surface topology. We first find all boundary vertex pairs satisfying a prescribed geometric condition in the areas between neighboring surfaces, and then separates areas to several regions by using boundary vertex pairs. We next compute best fitting planes suitable to each regions through PCA(Principal Component Analysis). They are used to produce triangles that be inserted into empty areas between neighboring surfaces. Since each regions between neighboring surfaces can be integrated by using local surface topology, a proposed method is robust to registration errors and surface scanning noise. We also propose a method integrating of textures by using parameterization technique. We first transforms integrated surface into initial viewpoints of each surfaces. We then project each textures to transformed integrated surface. They will be then assigned into parameter domain for integrated surface and be integrated according to the seaming lines for surfaces. Experimental results show that the proposed method is efficient to face modeling.

• Journal of Korea Multimedia Society
• /
• v.15 no.10
• /
• pp.1196-1204
• /
• 2012

In this paper, a series of algorithms are proposed for controlling different kinds of multimedia contents and realizing interact between human and computer by using single input device. Human gesture recognition based on NUI is presented firstly in my paper. Since the image information we get it from camera is not sensitive for further processing, we transform it to YCbCr color space, and then morphological processing algorithm is used to delete unuseful noise. Boundary Energy and depth information is extracted for hand detection. After we receive the image of hand detection, PCA algorithm is used to recognize hand posture, difference image and moment method are used to detect hand centroid and extract trajectory of hand movement. 8 direction codes are defined for quantifying gesture trajectory, so the symbol value will be affirmed. Furthermore, HMM algorithm is used for hand gesture recognition based on the symbol value. According to series of methods we presented, we can control multimedia contents by using human gesture recognition. Through large numbers of experiments, the algorithms we presented have satisfying performance, hand detection rate is up to 94.25%, gesture recognition rate exceed 92.6%, hand posture recognition rate can achieve 85.86%, and face detection rate is up to 89.58%. According to these experiment results, we can control many kinds of multimedia contents on computer effectively, such as video player, MP3, e-book and so on.

• The KIPS Transactions:PartD
• /
• v.14D no.4 s.114
• /
• pp.381-388
• /
• 2007

Consumers changed to the subject to produce a digital contents as data transmission technique is advanced and a digital machine is diffused variously. Users are interested greatly in a user creative movie (UCM) production among various online contents. The UCM production method which uses the music and picture is the method that users make the UCM more easily. However, the UCM production service has the problem that any association does not exist in the music and picture and that the picture changes according to fixed time interval without the relation at a music rhythm. To solve this problem, we propose the UCM production method which uses a music analysis and picture analysis in the paper. A music analysis finds a picture change time according to the rhythm and a picture analysis finds the association of the picture. A music analysis finds strong parts of the sound which uses Root-Mean-Square (RMS). And a picture analysis classifies the picture as a scenery picture and people picture which uses structure simplicity of the picture(SSP) and face region detection. A picture analysis got correct result of 86.4% in the experiment and we can finds the association at each picture and arranges the sequence which the picture appears. Therefore, if we use a music and picture analysis at the UCM production, users may make natural and efficient movie.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.2 s.308
• /
• pp.19-29
• /
• 2006

As the security of personal information is becoming more important in mobile phones, we are starting to apply iris recognition technology to these devices. In conventional iris recognition, magnified iris images are required. For that, it has been necessary to use large magnified zoom & focus lens camera to capture images, but due to the requirement about low size and cost of mobile phones, the zoom & focus lens are difficult to be used. However, with rapid developments and multimedia convergence trends in mobile phones, more and more companies have built mega-pixel cameras into their mobile phones. These devices make it possible to capture a magnified iris image without zoom & focus lens. Although facial images are captured far away from the user using a mega-pixel camera, the captured iris region possesses sufficient pixel information for iris recognition. However, in this case, the eye region should be detected for accurate iris recognition in facial images. So, we propose a new fast iris detection method, which is appropriate for mobile phones based on corneal specular reflection. To detect specular reflection robustly, we propose the theoretical background of estimating the size and brightness of specular reflection based on eye, camera and illuminator models. In addition, we use the successive On/Off scheme of the illuminator to detect the optical/motion blurring and sunlight effect on input image. Experimental results show that total processing time(detecting iris region) is on average 65ms on a Samsung SCH-S2300 (with 150MHz ARM 9 CPU) mobile phone. The rate of correct iris detection is 99% (about indoor images) and 98.5% (about outdoor images).

• The Journal of the Korea Contents Association
• /
• v.10 no.11
• /
• pp.20-29
• /
• 2010

The typical Kinetic Family Drawing (KFD) Web database system, a form of prototype system, has been developed, relying on the suggestions from family art therapists, with an aim to handle large amounts of assessment data and to facilitate effective implement of assessment activities. However, unfortunately such a system has an intrinsic problem that it fails to collect clients' behaviors, attitudes, facial expressions, voices, and other critical information observed while they are drawing. Accordingly we propose the ontology based video event analysis and video retrieval system in this paper, in order to enhance the function of a KFD Web database system by using a web camera and drawing tool. More specifically, a newly proposed system is designed to deliver two kinds of services: the client video retrieval service and the sketch video retrieval service, accompanied by a summary report of occurred events and dynamic behaviors relative to each family member object, respectively. The proposed system can support the reinforced KFD assessments by providing quantitative and subjective information on clients' working attitudes and behaviors, and KFD preparation processes.

• Science of Emotion and Sensibility
• /
• v.19 no.2
• /
• pp.3-18
• /
• 2016

This research, through video-based communication in a social video platform environment, studied the influence of the relationship between a video-watching subject and other watchers to that of the user's positive emotion and sense of community. Based on prior psychological theories called Social Impact Theory and Self-Monitoring Theory, the research built an actual video-based social video platform environment in order to verify an alternative utilizing new means of interaction based on videos. The result shows that under video-watching settings, user feels greater positive emotion and sense of community when the screen shows other people's reaction live and when him or her self's face is shown together, compared to when they are not shown. Also, based on the ANOVA analysis, the percentage of increase in positive emotion was greater when the two conditions mentioned above were provided synchronously compared to when they were not. The result of the research is expected to yield insights about a new form of social video platform.