• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.19 no.37
• /
• pp.41-51
• /
• 1996

This paper evaluates the clustering performance of a neural network and a statistical method. Algorithms which are used in this paper are the GLVQ(Generalized Learning vector Quantization) for a neural method and the k-means algorithm fer a statistical clustering method. For comparison of two methods, we calculate the Rand's c statistics. As a result, the mean of c value obtained with the GLVQ is higher than that obtained with the k-means algorithm, while standard deviation of c value is lower. Experimental data sets were the Fisher's IRIS data and patterns extracted from handwritten numerals.

• Proceedings of the KIEE Conference
• /
• 2006.04a
• /
• pp.114-116
• /
• 2006

This paper is to propose the palm print recognition system using wavelet transform. The palm print is frequently used as the material for the biometric recognition system such as the finger print, iris, face, etc. Since the palm print has lots of properties which include principle line, wrinkles, ridge and so forth, the ways of the implementation of the system are various. In this paper, at first, the palm print image is acquired and then some level of wavelet transform is performed. The coefficients become to be some blocks size of M by N after divided into the horizontal, vertical, diagonal components each level. The mean values, which are calculated with values of each block, are used as the feature vector. To compare between the stored template and the acquired vectors, we adopt the PNN (Probability Neural Network) method.

• The Journal of the Acoustical Society of Korea
• /
• v.17 no.4E
• /
• pp.3-10
• /
• 1998

This paper concerns the dynamical behavior, in probabilistic sense, of a feedforward neural network performing auto association for novelty. Networks of retinotopic topology having a one-to-one correspondence between and output units can be readily trained using back-propagation algorithm, to perform autoassociative mappings. A novelty filter is obtained by subtracting the network output from the input vector. Then the presentation of a "familiar" pattern tends to evoke a null response ; but any anomalous component is enhanced. Such a behavior exhibits a promising feature for enhancement of weak signals in additive noise. As an analysis of the novelty filtering, this paper shows that the probability density function of the weigh converges to Gaussian when the input time series is statistically characterized by nonsymmetrical probability density functions. After output units are locally linearized, the recursive relation for updating the weight of the neural network is converted into a first-order random differential equation. Based on this equation it is shown that the probability density function of the weight satisfies the Fokker-Planck equation. By solving the Fokker-Planck equation, it is found that the weight is Gaussian distributed with time dependent mean and variance.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.250-253
• /
• 2009

This paper presents NR (No Reference) Quality assessment method for IPTV or mobile IPTV. Because No Reference quality assessment method does not access the original signal so it is suitable for the real-time streaming service. Our proposed method use decoding parameters, such as quantization parameter, motion vector, and packet loss as a major network parameter. To evaluate performance of the proposed algorithm, we carried out subjective test of video quality with the ITU-T P.910 ACR (Absolute Category Rating) method and obtained the mean opinion score (MOS) value for QVGA 180 video sequence coded by H.264/AVC encoder. Experimental results show the proposed quality metric has a high correlation (84%) to subjective quality.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.408-412
• /
• 2009

In this paper, a new efficient algorithm for global motion estimation is proposed. This algorithm uses a previous 4-parameter model based global motion estimation algorithm and M-estimator for improving the accuracy and robustness of the estimate. The first algorithm uses the block based motion vector fields and which generates a coarse global motion parameters. And second algorithm is M-estimator technique for getting precise global motion parameters. This technique does not increase the computational complexity significantly, while providing good results in terms of estimation accuracy. In this work, an initial estimation for the global motion parameters is obtained using simple 4-parameter global motion estimation approach. The parameters are then refined using M-estimator technique. This combined algorithm shows significant reduction in mean compensation error and shows performance improvement over simple 4-parameter global motion estimation approach.

• Broadcasting and Media Magazine
• /
• v.22 no.1
• /
• pp.10-18
• /
• 2017

본 연구는 비접촉식 센서 기반의 웨어러블 디바이스를 이용한 딥러닝 기반의 제스처 인식에 대한 연구이다. 이를 위하여 Flexible MSG 센서를 기반으로 한 Flexible Epidermal Tactile Sensor를 사용하였으며, Flexible Epidermal Tactile Sensor는 손, 손가락 제스처를 취했을 때 손목, 손가락과 연결되어 있는 근육들의 움직임에 따라 발생하는 피부 표면의 전극을 취득하는 센서이다. 실험을 위하여 7가지 손, 손가락 제스처를 정의하였으며, 손목의 꺾임, 손목의 뒤틀림, 손가락의 오므림과 펴짐, 아무 동작도 취하지 않은 기본 상태에 대한 제스처로 정의하였다. 실험 데이터 수집에는 손목이나 손가락에 부상, 장애등이 없는 일반적인 8명의 참가자가 참가하였으며 각각 한 제스처에 대하여 20번씩 반복하여 1120개의 샘플을 수집하였다. 입력신호에 대한 제스처를 학습하기 위해 본 논문에서는 1차원 Convolutional Neural Network를 제안하였으며, 성능 비교를 위해 신호의 크기를 반영하는 특징벡터인 Integral Absolute Value와 Difference Absolute Mean Value를 입력신호에서 추출하고 Support Vector Machine을 사용하여 본 논문에서 제안한 1차원 CNN과 성능비교를 하였다. 그 결과 본 논문에서 제안한 1차원 CNN의 분류 정확도가 우수한 성능을 나타냈다.

• Proceedings of the KIPE Conference
• /
• 1998.10a
• /
• pp.298-303
• /
• 1998

In the speed sensorless control of the induction motor, the machine parameters (especially the secondary resistance R2) have a strong influence to the speed estimation. It is known that the simultaneous estimation of the speed and R2 is impossible in the slip frequency type vector control, because the secondary flux is constant. But the secondary flux is not always constant in the speed transient state. In this paper the R2 estimation in the transient state without adding any additional signal to the stator current is proposed. This algorithm uses the least mean square algorithm and the adaptive algorithm, and it is possible to estimate the R2 exactly. This algorithm is verified by the digital simulations and the experiments.

• Proceedings of the KIEE Conference
• /
• 2002.11c
• /
• pp.581-584
• /
• 2002

In this study, an algorithm for dithering true color image to 8-bit indexded color image using Artificial Neural Network was proposed. An adaptive vector quantization algorithm based on Artificial neural network was proposed for dithering color images. To evaluate the proposed algorithm, Mean Square Error(MSE) and quality between original image and dithered image was compared to those of other algorithm. As a results, MSE of proposed algorithm was lower than that of other algorithm used in commercial application and quality of dithered image was also highly improved.

• Management Science and Financial Engineering
• /
• v.5 no.1
• /
• pp.51-71
• /
• 1999

Performance of the group cumulative sum (CUSUM) control scheme using multiple univariate CUSUM charts is more sensitive to the change of quality control (QC) characteristics than the control chart schemes based on the Hotelling statistic We vexamine three group charts for multivariate normal data sets simulated with various correlation structures and shift directions in the mean vector. These group schemes apply the original measurement vectors, the scaled residual vectors from the re-gression of each variable on all others and the principal component vectors respectively to calculat-ing the CUSUM statistics. They are also compared to the multivariate QC charts based on the Ho-telling statistic by estimating average run lengths, coefficients of variation of run length and ranks in signaling order. On the basis of simulation results, we suggest a control chart scheme appropriate for specific quality control environment.

• Proceedings of the KOSOMBE Conference
• /
• v.1992 no.11
• /
• pp.29-33
• /
• 1992

We examine the compression performance of two irreversible (lossy) compression techniques, ADCT-VQ (Adaptive Discrete Cosine Trandform - Vector Quantization) and JPEG (Joint Photographic Experts group) which are basis of medical image information systems. Under the same compression ratio, MSE(Mean Square Error) is 0.578 lower in JPEG than in ADCT-VQ while SNR(Signal to Noise Ratio) is 1.236 dB higher in JPEG than in ADCT-VQ.