• Journal of IKEEE
• /
• v.21 no.4
• /
• pp.388-396
• /
• 2017

Since there are many variables such as various poses, illuminations and occlusions in a face detection problem, a high performance detection system is required. Although CNN is excellent in image classification, CNN operatioin requires high-performance hardware resources. But low cost low power environments are essential for small and mobile systems. So in this paper, the CPU-FPGA integrated system is designed based on 3-stage cascade CNN architecture using small size FPGA. Adaptive Region of Interest (ROI) is applied to reduce the number of CNN operations using face information of the previous frame. We use a Field Programmable Gate Array(FPGA) to accelerate the CNN computations. The accelerator reads multiple featuremap at once on the FPGA and performs a Multiply-Accumulate (MAC) operation in parallel for convolution operation. The system is implemented on Altera Cyclone V FPGA in which ARM Cortex A-9 and on-chip SRAM are embedded. The system runs at 30FPS with HD resolution input images. The CPU-FPGA integrated system showed 8.5 times of the power efficiency compared to systems using CPU only.

• Archives of design research
• /
• v.17 no.3
• /
• pp.39-50
• /
• 2004

According to the survey done among the 20s of university students in Oct. 2002, the female animation characters from Korea and Japan have almost no distinction. Speaking of the features of the characters, they have big eyes, a long nose, an oval face. However, their eyes and mouths are expressed exaggeratedly, which makes them hard to get a good impression from the people. It seems that the identity of Korean female animation characters is disappearing in terms of the copycat from western animation characters. Through the research, the features of 20s of Korean and Japanese girls are compared and they were reviewed from the ideal and biological point of view. Then, the animation characters, which physiognomy has been applied, are created based on result, which I am trying to find the standard model of Korea female animation character. As an ideal point of view, it turns out that Korean has a positive reaction on the female animation characters with the cute & introspective characteristics. On the contrary, Korean has a negative reactions on the ones that were created in a biological viewpoint and has a highly sensitive and discontented characteristics. This shows that Korean do not prefer the female animation characters with highly sensitive and discontented characters. According to the survey, 73.5% of people could find the same 7 kinds of female characters which are shown frequently in the animation. This means that the oriental physiognomy is an effective way to approach to the features of the character's face depending on the characteristics of each female characters. The characters in this study are based on real face of female, but we can create the various animation characters if we apply the partial feature of distinctive each personal face.

• Smart Media Journal
• /
• v.10 no.2
• /
• pp.76-83
• /
• 2021

With the development and use of smart devices such as smartphones and tablets increases, the mobile application market based on mobile devices is growing rapidly. Mobile application users write reviews to share their experience in using the application, which can identify consumers' various needs and application developers can receive useful feedback on improving the application through reviews written by consumers. However, there is a need to come up with measures to minimize the amount of time and expense that consumers have to pay to manually analyze the large amount of reviews they leave. In this work, we propose to collect delivery application user reviews from Google PlayStore and then use machine learning and deep learning techniques to classify them into four categories like application feature advantages, disadvantages, feature improvement requests and bug report. In the case of the performance of the Hugging Face's pretrained BERT-based Transformer model, the f1 score values for the above four categories were 0.93, 0.51, 0.76, and 0.83, respectively, showing superior performance than LSTM and GRU.

• Journal of KIISE:Computer Systems and Theory
• /
• v.32 no.10
• /
• pp.520-529
• /
• 2005

A feature-based 3D modeling algorithm is presented in this paper. Since conventional methods use depth-based techniques, they need much time for the image matching to extract depth information. Even feature-based methods have less computation load than that of depth-based ones, the calculation of modeling error about whole pixels within a triangle is needed in feature-based algorithms. It also increase the computation time. Therefore, the proposed algorithm consists of three phases, which are an initial 3D model generation, model evaluation, and model refinement phases, in order to acquire an efficient 3D model. Intensity gradients and incremental Delaunay triangulation are used in the Initial model generation. In this phase, a morphological edge operator is adopted for a fast edge filtering, and the incremental Delaunay triangulation is modified to decrease the computation time by avoiding the calculation errors of whole pixels and selecting a vertex at the near of the centroid within the previous triangle. After the model generation, sparse vertices are matched, then the faces are evaluated with the size, approximation error, and disparity fluctuation of the face in evaluation stage. Thereafter, the faces which have a large error are selectively refined into smaller faces. Experimental results showed that the proposed algorithm could acquire an adaptive model with less modeling errors for both smooth and abrupt areas and could remarkably reduce the model acquisition time.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.39 no.3
• /
• pp.23-35
• /
• 2002

Gaze detection is to locate the position on a monitor screen where a user is looking. In our work, we implement it with a computer vision system setting a single camera above a monitor and a user moves (rotates and/or translates) his face to gaze at a different position on the monitor. To detect the gaze position, we locate facial region and facial features(both eyes, nostrils and lip corners) automatically in 2D camera images. From the movement of feature points detected in starting images, we can compute the initial 3D positions of those features by camera calibration and parameter estimation algorithm. Then, when a user moves(rotates and/or translates) his face in order to gaze at one position on a monitor, the moved 3D positions of those features can be computed from 3D rotation and translation estimation and affine transform. Finally, the gaze position on a monitor is computed from the normal vector of the plane determined by those moved 3D positions of features. As experimental results, we can obtain the gaze position on a monitor(19inches) and the gaze position accuracy between the computed positions and the real ones is about 2.01 inches of RMS error.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.5 no.3
• /
• pp.427-433
• /
• 2001

This paper proposes a web-based 3 dimensional face modeling system that makes a realistic facial model efficiently without any 30 scanner or camera that uses in the traditional methods. Without expensive image-input equipments, we can easily create 3B models only using front and side images. The system is available to make 3D facial models as we connect to the facial modeling server on the WWW which is independent from specific platforms and softwares. This system will be implemented using Java 3D API, which includes the functions and conveniences of developed graphic libraries. It is a Client/server architecture which consists of user connection module and 3D facial model creating module. Clients connect with the facial modeling server, input two facial photographic images, detects the feature points, and then create a 3D facial model modifying generic facial model with the points according to the procedures using only the web browser.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.4
• /
• pp.431-437
• /
• 2004

This paper propose a facial recognition method based on an ubiquitous computing that is one of next generation intelligence technology fields. The facial images are acquired by a mobile device so-called cellular phone camera. We consider a mobile security using facial feature extraction and recognition process. Facial recognition is performed by the PCA and fuzzy LDA algorithm. Applying the discrete wavelet based on multi-resolution analysis, we compress the image data for mobile system environment. Euclidean metric is applied to measure the similarity among acquired features and then obtain the recognition rate. Finally we use the mobile equipment to show the efficiency of method. From various experiments, we find that our proposed method shows better results, even though the resolution of mobile camera is lower than conventional camera.

• Journal of the Korean Crystal Growth and Crystal Technology
• /
• v.27 no.4
• /
• pp.191-195
• /
• 2017

The microstructures of nonwoven fabric with and without nanofibers used as a filter for air purification type mask were investigated using scanning electron microscope. Moreover, we proposed a new mask design which is effective in the improvement of the fine dust blocking property. When comparing to nonwoven fabrics of which an average diameter was $25{\mu}m$, the nanofibers formed by the electrospinning process had a tight mesh structure arranged irregularly with a relatively large specific surface area, which could be associated with their much smaller diameter ranging from 25 to 120 nm. Such a prominent structural feature at nanofibers led to mechanical adsorption of fine particles, resulting in enhancement of filtering behavior maintaining high permeability. In addition to the excellent performance of the mask filter, wearing the mask properly is expected to maximize the blocking property of fine dust. To meet such a requirement, a new mask design that can be closely attached to the face in order to effectively block fine dust entering the gap between the face and the mask.

• Earthquakes and Structures
• /
• v.14 no.2
• /
• pp.103-115
• /
• 2018

In this work, a simple but accurate hyperbolic plate theory for the free vibration analysis of functionally graded material (FGM) sandwich plates is developed. The significant feature of this formulation is that, in addition to including the shear deformation effect, it deals with only 3 unknowns as the classical plate theory (CPT), instead of 5 as in the well-known first shear deformation theory (FSDT) and higher-order shear deformation theory (HSDT). A shear correction factor is, therefore, not required. Two common types of FGM sandwich plates are considered, namely, the sandwich with the FGM face sheet and the homogeneous core and the sandwich with the homogeneous face sheet and the FGM core. The equation of motion for the FGM sandwich plates is obtained based on Hamilton's principle. The closed form solutions are obtained by using the Navier technique. The fundamental frequencies are found by solving the eigenvalue problems. Numerical results of the present theory are compared with the CPT, FSDT, order shear deformation theories (HSDTs), and 3D solutions. Verification studies show that the proposed theory is not only accurate and simple in solving the free vibration behaviour of FGM sandwich plates, but also comparable with the higher-order shear deformation theories which contain more number of unknowns.

• Science of Emotion and Sensibility
• /
• v.5 no.4
• /
• pp.1-10
• /
• 2002

A web based virtual try-on provides customers a more enjoyable shopping experience that visualize clothes on personal mannequin. The researchers compared virtual fitting models which were developed in 2000 at Korea and in 2000 and 2002 at U.S. The results of this study as follows: The information about user's body size was required to input for selection of a virtual fitting model. 7 to 19 different body size, shape, and face features including weight and height were needed for visualizing virtual fitting model. The body type of the U.S virtual fitting model(My virtual model) was selected by front view silhouette for women, and by shoulder width and midriff silhouette for men. The more detailed information was required for developing Korean virtual fitting model. The additional body size information required in the site were leg and arm lengths, waist length, and thigh and ankle circumferences. The body proportion of Korean cyber personal mannequin was longer and narrower than the U.S cyber personal mannequin. It was recommended that standardized body length, width, and depth proportions calculated from national anthropometric data must be applied for developing Korean virtual fitting model. With application of more detailed information on face feature and advanced graphic image technology the 'My virtual model in 2002 resembled the human body shape of various race.