• Journal of IKEEE
• /
• v.24 no.1
• /
• pp.254-259
• /
• 2020

This paper is the implementation of the conversion system from image to music based on intentional synesthesia. The input image based on color, texture, and shape was converted into melodies, harmonies and rhythms of music, respectively. Depending on the histogram of colors, the melody can be selected and obtained probabilistically to form the melody. The texture in the image expressed harmony and minor key with 7 characteristics of GLCM, a statistical texture feature extraction method. Finally, the shape of the image was extracted from the edge image, and using Hough Transform, a frequency component analysis, the line components were detected to produce music by selecting the rhythm according to the distribution of angles.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.10a
• /
• pp.1045-1048
• /
• 2019

최근 CCTV 사용이 보편화되면서 방범 목적으로 서비스 시설이나 공공시설에 설치되는 CCTV의 수가 급격하게 증가하고 있다. 그에 따라 CCTV를 감시하는 노동력이 부족해지는 문제가 발생하여 이를 대체하기 위해 카메라 영상을 통하여 한번 인식한 사람을 다른 시간이나 장소에서 촬영된 영상에서 다시 인식하는 사람 재인식 기술이 주목받고 있다. 또한, 이러한 사람 재인식 기술은 보안 분야뿐만 아니라 영화나 드라마와 같은 영상 컨텐츠에 적용되어 불법 복제물을 찾는 일에 사용될 수도 있다. 기존의 사람 재인식에는 이미지의 유사도를 계산하는 방법이 사용되었지만, 조명이나 카메라 각도가 달라지면 성능이 급격하게 떨어지는 문제가 있었다. 최근에는 딥러닝 기술이 발달하면서 전반적인 영상처리 분야의 성능이 향상되었고, 사람 재인식 분야 역시 딥러닝을 활용하면서 성능이 향상되었다. 하지만 딥러닝을 활용한 방법의 경우 보통 두 개의 이미지를 입력으로 사용하여 같은지 다른지를 판단하게 되므로 각 이미지의 공통점이나 차이점을 동시에 고려하기는 어려운 점이 있다. 본 논문에서는 이러한 점을 해결하기 위해 세 개의 사람 이미지를 입력으로 사용하여 특징을 추출하고, 특징 맵을 재구성하여 각 이미지의 차이점과 공통점을 동시에 고려하며 학습할 수 있는 모델을 제안한다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.05a
• /
• pp.516-518
• /
• 2022

In this paper, we introduce a dependence of number of nodes of hidden-layer in fall detection system using Long Short-Term Memory that can detect falls. Its training is carried out using the parameter theta(θ), which indicates the angle formed by the x, y, and z-axis data for the direction of gravity using a 3-axis acceleration sensor. In its learning, validation is performed and divided into training data and test data in a ratio of 8:2, and training is performed by changing the number of nodes in the hidden layer to increase efficiency. When the number of nodes is 128, the best accuracy is shown with Accuracy = 99.82%, Specificity = 99.58%, and Sensitivity = 100%.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.294-296
• /
• 2022

Atherosclerosis is a chronic vascular inflammatory disease in which plaque builds up in the arteries and impairs blood flow. This can lead to heart disease and stroke. Since most people do not have any symptoms until the artery is severely narrowed, early detection of atherosclerosis is critical. In this paper, in order to effectively detect atherosclerotic lesions in tube-shaped blood vessels, polar conversion is applied to MRI images based on the vessel center. We then propose a SE-LSTMNet model using continuous signal information for each angle of a polar coordinate image. The trained model showed classification performance of 0.9194 accuracy, 0.9370 sensitivity, 0.8796 specificity, 0.8700 F1 score, and 0.9719 AUC on the validation data.

• Composites Research
• /
• v.36 no.1
• /
• pp.53-58
• /
• 2023

The PIC design method is assigning different stacking sequences for each shell element through the preliminary FE analysis. In previous study, machine learning was applied to the PIC design method in order to assign the region efficiently, and the training data is labeled by dividing each region into tension, compression, and shear through the preliminary FE analysis results value. However, since buckling is not considered, when buckling occurs, it can't be divided into appropriate loading type. In the present study, it was proposed PIC-NTL (PIC design using novel technique for analyzing load type) which is method for applying a novel technique for analyzing load type considering buckling to the conventional PIC design. The stress triaxiality for each ply were analyzed for buckling analysis, and the representative loading type was designated through the determined loading type within decision area divided into two regions of the same size in the thickness direction of the elements. The input value of the training data and label consisted in coordination of element and representative loading type of each decision area, respectively. A machine learning model was trained through the training data, and the hyperparameters that affect the performance of the machine learning model were tuned to optimal values through Bayesian algorithm. Among the tuned machine learning models, the SVM model showed the highest performance. Most effective stacking sequence were mapped into PIC tube based on trained SVM model. FE analysis results show the design method proposed in this study has superior external loading resistance and energy absorption compared to previous study.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.23 no.5
• /
• pp.423-430
• /
• 2013

In typical face recognition systems, the frontal view of face is preferred to reduce the complexity of the recognition. Thus individuals may be required to stare into the camera, or the camera should be located so that the frontal images are acquired easily. However these constraints severely restrict the adoption of face recognition to wide applications. To alleviate this problem, in this paper, we address the problem of tracking and recognizing faces in video captured with no environmental control. The face tracker extracts a sequence of the angle/size normalized face images using IVT (Incremental Visual Tracking) algorithm that is known to be robust to changes in appearance. Since no constraints have been imposed between the face direction and the video camera, there will be various poses in face images. Thus the pose is identified using a PCA (Principal Component Analysis)-based pose classifier, and only the pose-matched face images are used to identify person against the pre-built face DB with 5-poses. For face recognition, PCA, (2D)PCA, and $(2D)^2PCA$ algorithms have been tested to compute the recognition rate and the execution time.

• Geophysics and Geophysical Exploration
• /
• v.26 no.3
• /
• pp.126-137
• /
• 2023

In this study, a recurrent neural network (RNN) was employed as a methodological approach to classify dolphin click signals derived from ocean monitoring data. To improve the accuracy of click signal classification, the single time series data were transformed into fractional domains using fractional Fourier transform to expand its features. Transformed data were used as input for three RNN models: long short-term memory (LSTM), gated recurrent unit (GRU), and bidirectional LSTM (BiLSTM), which were compared to determine the optimal network for the classification of signals. Because the fractional Fourier transform displayed different characteristics depending on the chosen angle parameter, the optimal angle range for each RNN was first determined. To evaluate network performance, metrics such as accuracy, precision, recall, and F1-score were employed. Numerical experiments demonstrated that all three networks performed well, however, the BiLSTM network outperformed LSTM and GRU in terms of learning results. Furthermore, the BiLSTM network provided lower misclassification than the other networks and was deemed the most practically appliable to field data.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.11 no.8
• /
• pp.759-766
• /
• 2016

This paper is on a system design of recognizing sound sources and tracing locations from detecting a section of sound sources which is strong in surrounding environmental sounds about sound sources occurring in an abnormal situation by using signals within the section. In detection of the section with strong sound sources, weighted average delta energy of a short section is calculated from audio signals received. After inputting it into a low-pass filter, through comparison of values of the output result, a section strong in background sound is defined. In recognition of sound sources, from data of the detected section, using an HMM(: Hidden Markov Model) as a traditional recognition method, learning and recognition are realized from creating information to recognize sound sources. About signals of sound sources that surrounding background sounds are included, by using energy of existing signals, after detecting the section, compared with the recognition through the HMM, a recognition rate of 3.94% increase is shown. Also, based on the recognition result, location grasping by using TDOA(: Time Delay of Arrival) between signals in the section accords with 97.44% of angles of a real occurrence location.

• Journal of Korea Multimedia Society
• /
• v.15 no.5
• /
• pp.606-614
• /
• 2012

Face detection using AdaBoost algorithm is one of the fastest and the most robust face detection algorithm so many improvements or extensions of this method have been proposed. However, almost all previous approaches deal with only frontal face and suffer from limited discriminant capability for inclined face because these methods apply the same features for both frontal and inclined face. Also conventional approaches for detecting inclined face which apply frontal face detecting method to inclined input image or make different detectors for each angle require heavy computational complexity and show low detection rate. In order to overcome this problem, a method for detecting inclined face using JointBoost is proposed in this paper. The computational and sample complexity is reduced by finding common features that can be shared across the classes. Simulation results show that the detection rate of the proposed method is at least 2% higher than that of the conventional AdaBoost method under the learning condition with the same iteration number. Also the proposed method not only detects the existence of a face but also gives information about the inclined direction of the detected face.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.15 no.9
• /
• pp.1871-1880
• /
• 2011

In this paper, a controller for two wheeled inverted pendulum robot, i.e., Segway type robot that is a convenient and easily handled vehicle is designed to have more stable balancing and faster velocity control compared to the conventional method. First, a widely used PID control structure is applied to the two wheeled inverted pendulum robot and proper PID control gains for some specified weights of users are obtained to get accurate balancing and velocity control by use of experimental trial-and-error method. Next, neural network is employed to generate appropriate PID control gains for arbitrarily selected weight. Here the PID gains based on the trial-and-error method are used as training data. Simulation study has been carried out to find that the performance of the designed controller using the neural network is more excellent than the conventional PID controller in terms of faster balancing and velocity control.