• Journal of Korea Multimedia Society
• /
• v.3 no.6
• /
• pp.592-603
• /
• 2000

This paper is concerning off-line signature verification using a density function which is obtained by convolving the signature image with twelve-directional $5\times{5}$ gradient masks and the weighted fuzzy mean classifier. The twelve-directional density function based on Nevatia-Babu template gradient is related to the overall shape of a signature image and thus, utilized as a feature set. The weighted fuzzy mean classifier with the reference feature vectors extracted from only genuine signature samples is evaluated for the verification of freehand forgeries. The experimental results show that the proposed system can classify a signature whether it is genuine or forged with more than 98% overall accuracy even without any knowledge of varied freehand forgeries.

• Korean Journal of Cognitive Science
• /
• v.8 no.4
• /
• pp.55-62
• /
• 1997

We proposed a method for locating of mouth using deformable templates, described by a parameterized template. An energy function is defined which links, edges, peaks, valleys in image intensity to corresponding properties of the template. The template deforms itself by altering its parameter values to minimize the energy function. The minimized energy function's parameter values can be used as descriptors for the feature. We propose a method for locating mouth fast, accurately by limiting a range of parameters' value and getting initial value of parameters' by preprocessing.

• 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.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.9
• /
• pp.2882-2903
• /
• 2022

Image super-resolution (SR) processing is of great value in the fields of digital image processing, intelligent security, film and television production and so on. This paper proposed a densely connected deep learning network based on cascade architecture, which can be used to solve the problem of super-resolution in the field of image quality enhancement. We proposed a more efficient residual scaling dense block (RSDB) and the multi-channel cascade architecture to realize more efficient feature reuse. Also we proposed a hybrid loss function based on L₁ error and L_∞ error to achieve better L_∞ error performance. The experimental results show that the overall performance of the network is effectively improved on cascade architecture and residual scaling. Compared with the residual dense net (RDN), the PSNR / SSIM of the new method is improved by 2.24% / 1.44% respectively, and the L_∞ performance is improved by 3.64%. It shows that the cascade connection and residual scaling method can effectively realize feature reuse, improving the residual convergence speed and learning efficiency of our network. The L_∞ performance is improved by 11.09% with only a minimal loses of 1.14% / 0.60% on PSNR / SSIM performance after adopting the new loss function. That is to say, the L_∞ performance can be improved greatly on the new loss function with a minor loss of PSNR / SSIM performance, which is of great value in L_∞ error sensitive tasks.

• Smart Structures and Systems
• /
• v.33 no.4
• /
• pp.253-262
• /
• 2024

Current-based mechanical fault diagnosis is more convenient and low cost since additional sensors are not required. However, it is still challenging to achieve this goal due to the weak fault information in current signals. In this paper, a dual-loss convolutional neural network (DLCNN) is proposed to implement the intelligent bearing fault diagnosis via current signals. First, a novel similarity loss (SimL) function is developed, which is expected to maximize the intra-class similarity and minimize the inter-class similarity in the model optimization operation. In the loss function, a weight parameter is further introduced to achieve a balance and leverage the performance of SimL function. Second, the DLCNN model is constructed using the presented SimL and the cross-entropy loss. Finally, the two-phase current signals are fused and then fed into the DLCNN to provide more fault information. The proposed DLCNN is tested by experiment data, and the results confirm that the DLCNN achieves higher accuracy compared to the conventional CNN. Meanwhile, the feature visualization presents that the samples of different classes are separated well.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.28 no.3
• /
• pp.353-359
• /
• 2010

In this paper, we propose the automatic image-to-image registration of high resolution satellite images using local properties of tie points to improve the registration accuracy. A spatial distance between interest points of reference and sensed images extracted by Scale Invariant Feature Transform(SIFT) is additionally used to extract tie points. Coefficients of affine transform between images are extracted by invariant descriptor based matching, and interest points of sensed image are transformed to the reference coordinate system using these coefficients. The spatial distance between interest points of sensed image which have been transformed to the reference coordinates and interest points of reference image is calculated for secondary matching. The piecewise linear function is applied to the matched tie points for automatic registration of high resolution images. The proposed method can extract spatially well-distributed tie points compared with SIFT based method.

• Journal of the Korea Society of Computer and Information
• /
• v.12 no.1 s.45
• /
• pp.51-56
• /
• 2007

Feature functions were used for the classification of music. The spectral roll-off, variance, average peak level, and class were chosen to make up a feature function vector. Among these, it is the spectral roll-off function that has a low-frequency to high-frequency ratio. To find the optimal roll-off point, the roll-off points from 0.05 to 0.95 were swept. The classification success rate was monitored as the roll-off point was being changed. The data that were used for the experiments were taken from the sounds made by a modern violin and a baroque one. Their shapes and sounds are similar, but they differ slightly in sound texture. As such, the data obtained from the sounds of these two kinds of violin can be useful in finding an adequate roll-off point. The optimal roll-off point, as determined through the experiment, was 0.85. At this point, the classification success rate was 85%, which was the highest.

• Physical Therapy Korea
• /
• v.26 no.3
• /
• pp.42-56
• /
• 2019

Background: Individuals with mechanical neck pain show biomechanical and neurophysiological changes, including cervical spine muscle weakness. As a result of deep muscle weakness, it causes stability disability and reduced upper thoracic spine mobility, which finally leads to functional movement restriction such as limited range of motion and dysfunction. Recent studies have shown that thoracic spine manipulation and mobilization could reduce symptoms of mechanical neck pain in patients. Objects: The purpose of this study was to investigate the effects of thoracic mobility exercise on cervicothoracic function, posture feature, and pain intensity in individuals with mechanical neck pain. Methods: The study subjects were 26 persons who were randomly assigned to the experimental (with thoracic mobility exercise) and control groups (without thoracic mobility exercise), with 13 subjects in each group. The cervicothoracic function (neck functional disability level and cervicothoracic range of motion), posture feature, and pain rating (using a quadrupled visual analogue scale [QVAS]) were measured before, after 3 weeks, and after 6 weeks. Results: Statistically significant group-by-time interactions were found with repeated analyses of variance for the Korean neck disability index (KNDI), all cervical range of motion (CROM), all thoracic range of motion (TROM), cranial rotation angle, sagittal shoulder posture (SSP), and QVAS (p<.05). All groups showed significant improvements from all times in all the evaluated methods. The KNDI, CROM, TROM of left rotation, and SSP in the experimental group showed significant improvements after 3 weeks, and the TROM of the right rotation and QVAS in the experimental group showed significant improvements after 6 weeks when compared with the control group. Conclusion: Thoracic mobility exercise during 6 weeks might be effective intervention to improve the functional level, posture feature, and QVAS pain rating for managing individuals with mechanical neck pain.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.66 no.9
• /
• pp.1392-1401
• /
• 2017

In this study, the design methodology of pattern classification is introduced for avoiding faults through partial discharge occurring in the power facilities and local sites. In order to classify some partial discharge types according to the characteristics of each feature, the model is constructed by using the Radial Basis Function Neural Networks(RBFNNs) and Particle Swarm Optimization(PSO). In the input layer of the RBFNNs, the feature vector is searched and the dimension is reduced through Principal Component Analysis(PCA) and PSO. In the hidden layer, the fuzzy coefficients of the fuzzy clustering method(FCM) are tuned using PSO. Raw datasets for partial discharge are obtained through the Motor Insulation Monitoring System(MIMS) instrument using an Epoxy Mica Coupling(EMC) sensor. The preprocessed datasets for partial discharge are acquired through the Phase Resolved Partial Discharge Analysis(PRPDA) preprocessing algorithm to obtain partial discharge types such as void, corona, surface, and slot discharges. Also, when the amplitude size is considered as two types of both the maximum value and the average value in the process for extracting the preprocessed datasets, two different kinds of feature datasets are produced. In this study, the classification ratio between the proposed RBFNNs model and other classifiers is shown by using the two different kinds of feature datasets, and also we demonstrate the proposed model shows superiority from the viewpoint of classification performance.

• Journal of Korea Multimedia Society
• /
• v.13 no.8
• /
• pp.1115-1127
• /
• 2010

In order to increase the productivity in manufacturing TFT-LCD(thin film transistor-liquid crystal display), it is essential to classify defects that occur during the production and make an appropriate decision on whether the product with defects is scrapped or not. The decision mainly depends on classifying the defects accurately. In this paper, we present an effective classification method for film defects acquired in the panel production line by analyzing the intensity distribution and shape feature of the defects. We first generate a binary image for each defect by separating defect regions from background (non-defect) regions. Then, we extract various features from the defect regions such as the linearity of the defect, the intensity distribution, and the shape characteristics considering intensity, and construct a referential image database that stores those feature values. Finally, we determine the type of a defect by matching a defect image with a referential image in the database through the matching cost function between the two images. To verify the effectiveness of our method, we conducted a classification experiment using defect images acquired from real TFT-LCD production lines. Experimental results show that our method has achieved highly effective classification enough to be used in the production line.