• Journal of KIISE:Software and Applications
• /
• v.31 no.4
• /
• pp.403-410
• /
• 2004

Bimodal speech recognition systems have been proposed for enhancing recognition rate of ASR under noisy environments. Visual feature extraction is very important to develop these systems. To extract visual features, it is necessary to detect exact lip position. This paper proposed the method that detects a lip position using color similarity model and SVM. Face/Lip color distribution is teamed and the initial lip position is found by using that. The exact lip position is detected by scanning neighbor area with SVM. By experiments, it is shown that this method detects lip position exactly and fast.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2005.11a
• /
• pp.287-292
• /
• 2005

This paper presents a rational polynomial approximation method to estimate modal parameters of wind excited structures using incomplete noisy measurements of structural responses and partial measurements of wind velocities only. A stochastic model of the excitation wind force acting on the structure is estimated from partial measurements of wind velocities. Then the transfer functions of the structure are approximated as rational polynomial functions. From the poles and zeros of the estimated rational polynomial functions, the modal parameters, such as natural frequencies, damping ratios, and mode shapes are extracted. Since the frequency characteristics of wind forces acting on structures can be assumed as a smooth Gaussian process especially around the natural frequencies of the structures according to the central limit theorem (Brillinger, 1969; Yaglom, 1987), the estimated modal parameters are robust and reliable with respect to the assumed stochastic input models. To verify the proposed method, the modal parameters of a TV transmission tower excited by gust wind are estimated. Comparison study with the results of other researchers shows the efficacy of the suggested method.

• Proceedings of the Korean Information Science Society Conference
• /
• 2010.06c
• /
• pp.62-65
• /
• 2010

Image annotations allow users to access a large image database with textual queries. But since the surrounding text of Web images is generally noisy. an efficient image annotation and retrieval system is highly desired. which requires effective image search techniques. Data mining techniques can be adopted to de-noise and figure out salient terms or phrases from the search results. Clustering algorithms make it possible to represent visual features of images with finite symbols. Annotationbased image search engines can obtains thousands of images for a given query; but their results also consist of visually noise. In this paper. we present a new algorithm Double-Circles that allows a user to remove noise results and characterize more precise representative annotations. We demonstrate our approach on images collected from Flickr image search. Experiments conducted on real Web images show the effectiveness and efficiency of the proposed model.

• Korean Journal of Remote Sensing
• /
• v.19 no.6
• /
• pp.479-490
• /
• 2003

This study proposed a modified anisotropic diffusion restoration for image classification. The anisotropic diffusion restoration uses a probabilistic model based on Markov random field, which represents geographical connectedness existing in many remotely sensed images, and restores them through an iterative diffusion processing. In every iteration, the bonding-strength coefficient associated with the spatial connectedness is adaptively estimated as a function of brightness gradient. The gradient function involves a constant called "temperature", which determines the amount of discontinuity and is continuously decreased in the iterations. In this study, the proposed method has been extensively evaluated using simulated images that were generated from various patterns. These patterns represent the types of natural and artificial land-use. The simulated images were restored by the modified anisotropic diffusion technique, and then classified by a multistage hierarchical clustering classification. The classification results were compared to them of the non-restored simulation images. The restoration with an appropriate temperature considerably reduces error in classification, especially for noisy images. This study made experiments on the satellite images remotely sensed on the Korean peninsula. The experimental results show that the proposed approach is also very effective on image classification in remote sensing.

• Interaction and multiscale mechanics
• /
• v.3 no.2
• /
• pp.173-191
• /
• 2010

Damage detection plays a very important role to the maintenance of bridge structures. Traditional damage detection methods are usually based on structural dynamic properties, which are acquired from pre-installed sensors on the bridge. This is not only time-consuming and costly, but also suffers from poor sensitivity to damage if only natural frequencies and mode shapes are concerned in a noisy environment. Recently, the idea of using the dynamic responses of a passing vehicle shows a convenient and economical way for damage detection of bridge structures. Inspired by this new idea and the well-established tap test in the field of non-destructive testing, this paper proposes a new method for obtaining the damage information through the acceleration of a passing vehicle enhanced by a tapping device. Since no finger-print is required of the intact structure, this method can be easily implemented in practice. The logistics of this method is illustrated by a vehicle-bridge interaction model, along with the sensitivity analysis presented in detail. The validity of the method is proved by some numerical examples, and remarks are given concerning the potential implementation of the method as well as the directions for future research.

• International Journal of Computer Science & Network Security
• /
• v.21 no.12spc
• /
• pp.556-564
• /
• 2021

Numerous spatial and transform-domain-based conventional denoising algorithms struggle to keep critical and minute structural features of the image, especially at high noise levels. Although neural network approaches are effective, they are not always reliable since they demand a large quantity of training data, are computationally complicated, and take a long time to construct the model. A new framework of enhanced hybrid filtering is developed for denoising color images tainted by additive white Gaussian Noise with the goal of reducing algorithmic complexity and improving performance. In the first stage of the proposed approach, the noisy image is refined using a high-dimensional non-local means filter based on Principal Component Analysis, followed by the extraction of the method noise. The wavelet transform and SURE Shrink techniques are used to further culture this method noise. The final denoised image is created by combining the results of these two steps. Experiments were carried out on a set of standard color images corrupted by Gaussian noise with multiple standard deviations. Comparative analysis of empirical outcome indicates that the proposed method outperforms leading-edge denoising strategies in terms of consistency and performance while maintaining the visual quality. This algorithm ensures homogeneous noise reduction, which is almost independent of noise variations. The power of both the spatial and transform domains is harnessed in this multi realm consolidation technique. Rather than processing individual colors, it works directly on the multispectral image. Uses minimal resources and produces superior quality output in the optimal execution time.

• International Journal of Contents
• /
• v.17 no.4
• /
• pp.1-15
• /
• 2021

Named entity recognition (NER) is a crucial task for NLP, which aims to extract information from texts. To build NER systems, deep learning (DL) models are learned with dictionary features by mapping each word in the dataset to dictionary features and generating a unique index. However, this technique might generate noisy labels, which pose significant challenges for the NER task. In this paper, we proposed DL-dictionary features, and evaluated them on two datasets, including the OntoNotes 5.0 dataset and our new infectious disease outbreak dataset named GFID. We used (1) a Bidirectional Long Short-Term Memory (BiLSTM) character and (2) pre-trained embedding to concatenate with (3) our proposed features, named the Convolutional Neural Network (CNN), BiLSTM, and self-attention dictionaries, respectively. The combined features (1-3) were fed through BiLSTM - Conditional Random Field (CRF) to predict named entity classes as outputs. We compared these outputs with other predictions of the BiLSTM character, pre-trained embedding, and dictionary features from previous research, which used the exact matching and partial matching dictionary technique. The findings showed that the model employing our dictionary features outperformed other models that used existing dictionary features. We also computed the F1 score with the GFID dataset to apply this technique to extract medical or healthcare information.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.11a
• /
• pp.29-32
• /
• 2020

Due to the lack of improper image acquisition process, noise induction is an inevitable step. As a result, objective image quality assessment (IQA) plays an important role in estimating the visual quality of noisy image. Plenty of IQA methods have been proposed including traditional signal processing based methods as well as current deep learning based methods where the later one shows promising performance due to their complex representation ability. The deep learning based methods consists of several convolution layers and down sampling layers for feature extraction and fully connected layers for regression. Usually, the down sampling is performed by using max-pooling layer after each convolutional block. We reveal that this max-pooling causes information loss despite of knowing their importance. Consequently, we propose a better IQA method that replaces the max-pooling layers with strided convolutions to down sample the feature space and since the strided convolution layers have learnable parameters, they preserve optimal features and discard redundant information, thereby improve the prediction accuracy. The experimental results verify the effectiveness of the proposed method.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.3
• /
• pp.23-31
• /
• 2010

The increasing demands on low-power, and low-complexity video encoder have been motivating extensive research activities on distributed video coding (DVC) in which the encoder compresses frames without utilizing inter-frame statistical correlation. In DVC encoder, contrary to the conventional video encoder, an error control code compresses the video frames by representing the frames in the form of syndrome bits. In the meantime, the DVC decoder generates side information which is modeled as a noisy version of the original video frames, and a decoder of the error-control code corrects the errors in the side information with the syndrome bits. The noisy observation, i.e., the side information can be understood as the output of a virtual channel corresponding to the orignal video frames, and the conditional probability of the virtual channel model is assumed to follow a Laplacian distribution. Thus, performance improvement of DVC systems depends on performances of the error-control code and the optimal reconstruction step in the DVC decoder. In turn, the performances of two constituent blocks are directly related to a better estimation of the parameter of the correlation channel. In this paper, we propose an algorithm to estimate the parameter of the correlation channel and also a low-complexity version of the proposed algorithm. In particular, the proposed algorithm minimizes squared-error of the Laplacian probability distribution and the empirical observations. Finally, we show that the conventional algorithm can be improved by adopting a confidential window. The proposed algorithm results in PSNR gain up to 1.8 dB and 1.1 dB on Mother and Foreman video sequences, respectively.

• Korean Journal of Materials Research
• /
• v.9 no.1
• /
• pp.86-91
• /
• 1999

We have fabricated high-$\textrm{T}_c$ superconducting $\textrm{YBa}_{2}\textrm{Cu}_{3}\textrm{O}_{7-x}$(YBCO) grain boundary junctions at a step-edge on (001) $\textrm{SrTiO}_3$(STO) substrates. A diamond-like carbon (DLC) film grown by plasma enhanced chemical vapor deposition were used as an ion milling mask to make steps on the STO (100) single crystal and was removed by an oxygen reactive ion etch process. The c-axis oriented YBCO and TO thin films were deposited epitaxially on the STO substrate with a step-edge by pulsed laser deposition. The grain boundary junctions were formed at the top and the bottom of the step. The junctions worked at temperatures above 77 K, and had I\ulcornerR\ulcorner products of 7.5mV at 16K and 0.3 mV at 77K, respectively. The I-V characteristics of these junctions showed the shape of the two noisy resistively shunted junction model.