• Journal of Korea Multimedia Society
• /
• v.24 no.10
• /
• pp.1380-1390
• /
• 2021

The video face recognition (FR) is one of the most popular researches in the field of computer vision due to a variety of applications. In particular, research using the attention mechanism is being actively conducted. In video face recognition, attention represents where to focus on by using the input value of the whole or a specific region, or which frame to focus on when there are many frames. In this paper, we propose a novel attention based deep learning method. Main novelties of our method are (1) the use of combining two loss functions, namely weighted Softmax loss function and a Triplet loss function and (2) the feasibility of end-to-end learning which includes the feature embedding network and attention weight computation. The feature embedding network has a positive effect on the attention weight computation by using combined loss function and end-to-end learning. To demonstrate the effectiveness of our proposed method, extensive and comparative experiments have been carried out to evaluate our method on IJB-A dataset with their standard evaluation protocols. Our proposed method represented better or comparable recognition rate compared to other state-of-the-art video FR methods.

• The KSFM Journal of Fluid Machinery
• /
• v.11 no.1
• /
• pp.52-60
• /
• 2008

Staggered dimples printed on opposite walls of an internal cooling channel are formulated numerically and optimized to enhance heat transfer performance. Nusselt number and friction factor based objectives are considered and a weighted average surrogate model is used to approximate the data generated by numerical simulation. The dimpled channel shape is defined by three geometric design variables, and the design point within design space are selected using Latin hypercube sampling. A weighted-sum method for multi-objective optimization is applied to integrate multiple objectives into a single objective. By the optimization, the objective function value is improved largely and heat transfer rate is increase much higher than pressure loss increase due to shape deformation. Channel with vertically non-symmetric optimum dimples is tested and found that the best appears if dimples on opposite wall are displaced by one quarter of dimple spacing.

• International Journal of Naval Architecture and Ocean Engineering
• /
• v.12 no.1
• /
• pp.617-635
• /
• 2020

The purpose of this study is to discuss how to improve the maneuverability of lifting and diving for underwater vehicle's vertical motion. Therefore, to solve these problems, applied the 3-D numerical simulation, Taguchi's Design of Experiment (DOE), and intelligent parameter design methods, etc. We planned four steps as follows: firstly, we applied the 2-D flow simulation with NACA series, and then through the Taguchi's dynamic method to analyze the sensitivity (β). Secondly, take the data of pitching torque and total resistance from the Taguchi orthogonal array (L9), the ignal-to-noise ratio (SNR), and analysis each factorial contribution by ANOVA. Thirdly, used Radial Basis Function Network (RBFN) method to train the non-linear meta-modeling and found out the best factorial combination by Particle Swarm Optimization (PSO) and Weighted Percentage Reduction of Quality Loss (WPRQL). Finally, the application of the above methods gives the global optimum for multi-quality characteristics and the robust design configuration, including L/D is 9.4:1, the foreplane on the hull (Bow-2), and position of the sail is 0.25 Ls from the bow. The result shows that the total quality is improved by 86.03% in comparison with the original design.

• The Journal of the Acoustical Society of Korea
• /
• v.18 no.4
• /
• pp.23-30
• /
• 1999

In this study, we propose a new optimization algorithm for a withdrawal weighted SAW transversal filter to satisfy given, specifications such as bandwidth, ripple, insertion loss, and sidelobe rejection level. An analysis tool for the withdrawal weighted filter has been produced by means of the delta function model, and has been applied to the design of a filter consisting of an uniform input IDT and a withdrawal weighted output IDT. This optimization algorithm consists of three routines, which eventually determines eight design parameters to satisfy the performance specifications. At the first step, the number of input and output IDT fingers and their geometrical size are determined by the insertion loss specification. At the next step, the bandwidth is controlled by the change of the IDT finger position. Finally, the sidelobe rejection level is modified through the add/skip technique of IDT fingers. The algorithm in this paper is distinct from conventional techniques in that it can simultaneously consider all the specifications such as bandwidth, ripple, sidelobe rejection level and insertion loss, and optimize the geometry of the withdrawal weighted SAW filter.

• Journal of the Korean Professional Engineers Association
• /
• v.22 no.2
• /
• pp.19-25
• /
• 1989

Utilizing a cosine transform in image compression has several recognized performance benefits, resulting in the ability to attain large compression ratio with small quality loss. Also, various models incorporating Human Visual System (HVS) to Discrete Cosine Trans-form (DCT) scheme are considered. Using the exact frequency components of DCT basis function, the optimum modulation transfer function (MTF) is obtained analytically. The errors at a block boundary which is important factor in transform coder are criteria for error measurement. The HVS weight coding results in perceptually higher quality images compared with the unweighted scheme.

• Communications for Statistical Applications and Methods
• /
• v.20 no.6
• /
• pp.481-490
• /
• 2013

Classification is an important research area as data can be easily obtained even if the number of predictors becomes huge. The support vector machine(SVM) is widely used to classify a subject into a predetermined group because it gives sound theoretical background and better performance than other methods in many applications. The SVM can be viewed as a penalized method with the hinge loss function and penalty functions. Instead of $L_2$ penalty function Fan and Li (2001) proposed the smoothly clipped absolute deviation(SCAD) satisfying good statistical properties. Despite the ability of SVMs, they have drawbacks of non-robustness when there are outliers in the data. We develop a robust SVM method using a weight function with the SCAD penalty function based on the local quadratic approximation. We compare the performance of the proposed SVM with the SVM using the $L_1$ and $L_2$ penalty functions.

• IE interfaces
• /
• v.24 no.2
• /
• pp.151-155
• /
• 2011

In this study, we propose a signomial classification method with ₀-regularization (₀-)which seeks a sparse signomial function by solving a mixed-integer program to minimize the weighted sum of the ₀-norm of the coefficient vector of the resulting function and the $L_1$-norm of loss caused by the function. $SC_0$ gives an explicit description of the resulting function with a small number of terms in the original input space, which can be used for prediction purposes as well as interpretation purposes. We present a practical implementation of $SC_0$ based on the mixed-integer programming and the column generation procedure previously proposed for the signomial classification method with $SL_1$-regularization. Computational study shows that $SC_0$ gives competitive performance compared to other widely used learning methods for classification.

• Progress in Medical Physics
• /
• v.9 no.4
• /
• pp.217-225
• /
• 1998

In this paper, the analysis of the relation between noise induced hearing loss and noise exposure is described for the A-weighted noise levels and exposure duration. The hearing loss and sensitivity threshold shift is investigated by changing the various parameters such as the effects of aging and noise exposure. Total hearing loss is proportional to a function of exposure level based upon the integral of pressure with time. If the noise exposure term is large so that the presbycusis may be neglected, the hearing loss due to aging and noise exposure becomes the noise induced hearing loss. It is shown that exposure to excessive noise can cause temporary loss of hearing that may become permanent if the exposure is prolonged or intense. An audiogram taken from a person suffering from noise induced hearing loss will usually show the greatest loss of hearing sensitivity in the 4kHz region, which is typically the region most sensitive to damage resulting from many types of industrial noise.

• International Journal of Fluid Machinery and Systems
• /
• v.4 no.1
• /
• pp.172-178
• /
• 2011

This paper presents an optimization procedure for high-efficiency design of a mixed-flow pump. Optimization techniques based on a weighted-average surrogate model are used to optimize a vane diffuser of a mixed-flow pump. Validation of the numerical results is performed through experimental data for head, power and efficiency. Three-level full factorial design is used to generate nine design points within the design space. Three-dimensional Reynoldsaveraged Navier-Stokes equations with the shear stress transport turbulence model are discretized by using finite volume approximation and solved on hexahedral grids to evaluate the efficiency as the objective function. In order to reduce pressure loss in the vane diffuser, two variables defining the straight vane length ratio and the diffusion area ratio are selected as design variables in the present optimization. As the results of the design optimization, the efficiency at the design flow coefficient is improved by 7.05% and the off-design efficiencies are also improved in comparison with the reference design.

• ETRI Journal
• /
• v.41 no.4
• /
• pp.415-425
• /
• 2019

Generative adversarial networks (GAN) have been successfully used in many image restoration tasks, including image denoising, super-resolution, and compression artifact reduction. By fully exploiting its characteristics, state-of-the-art image restoration techniques can be used to generate images with photorealistic details. However, there are many applications that require faithful rather than visually appealing image reconstruction, such as medical imaging, surveillance, and video coding. We found that previous GAN-training methods that used a loss function in the form of a weighted sum of fidelity and adversarial loss fails to reduce fidelity loss. This results in non-negligible degradation of the objective image quality, including peak signal-to-noise ratio. Our approach is to alternate between fidelity and adversarial loss in a way that the minimization of adversarial loss does not deteriorate the fidelity. Experimental results on compression-artifact reduction and super-resolution tasks show that the proposed method can perform faithful and photorealistic image restoration.