• 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 KIEE Conference
• /
• 1987.07b
• /
• pp.1033-1037
• /
• 1987

This paper proposes a basic implementation of bi-directional transition system between Korean and Japanese. The syntactic structure of Korean language is very similar to that of Japanese. In order to use the same main program for bi-directional translation, system is classified into two parts. One is directly translatable part and the other is untranslatable part by direct matching because of different structure. Solving these problems, the syntactic and semantic informations control bi-directional interpreting process and make equivalent syntactic level between two languages.

• Journal of the Optical Society of Korea
• /
• v.19 no.5
• /
• pp.537-543
• /
• 2015

Liquid crystal displays (LCDs) have slow responses, so motion blurs are often perceived in fast moving scenes. To reduce this motion blur, we propose a novel method of robust motion compensated frame interpolation (MCFI) based on bidirectional motion estimation (BME) and weight-overlapped block motion compensation (WOBMC) with variable block sizes. In most MCFI methods, a static block size is used, so some block artefacts and motion blurs are observed. However, the proposed method adjusts motion block sizes and search ranges by comparing matching scores, so the precise motion vectors can be estimated in accordance with motions. In the MCFI, overlapping ranges for WOBMC are also determined by adjusted block sizes, so the accurate MCFI can be performed. In the experimental results, the proposed method strongly reduced motion blurs arisen from large motions, and yielded interpolated images with high visual performance and peak signal-to-noise ratio (PSNR).

• Proceedings of the IEEK Conference
• /
• 2006.06a
• /
• pp.437-438
• /
• 2006

Virtual view synthesis is essential for 3DTV systems, which utilizes the motion parallax cue. In this paper, we propose a multi-step view synthesis algorithm to efficiently reconstruct an arbitrary view from limited number of known views of a 3D scene. We describe an efficient image rectification procedure which guarantees that an interpolation process produce valid views. This rectification method can deal with all possible camera motions. The idea consists of using a polar parameterization of the image around the epipole. Then, to generate intermediate views, we use an efficient dense disparity estimation algorithm considering features of stereo image pairs. Main concepts of the algorithm are based on the region dividing bidirectional pixel matching. The estimated disparities are used to synthesize intermediate view of stereo images. We use computer simulation to show the result of the proposed algorithm.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2002.11a
• /
• pp.727-730
• /
• 2002

In this paper. we propose an algorithm to automatically generate a novel scene along the alternated viewing positions using the bidirectional disparity morphing. To compute the disparity between two reference images, we use the 2-step block based matching algorithm and define three occluding patterns so as to smooth the computed disparities, especially for occluded regions. We discuss the advantages of this algorithm compared to the commonly used schemes and show some experimental results with real data.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.6
• /
• pp.123-131
• /
• 2009

The emerging display technology has been replaced the previous position of the CRT with the LCD. The nature of hold type display such as LCD, however, causes many problems such as motion blur and motion judder. To resolve the problems, we used frame interpolation technique which improves the image quality by inserting new interpolated frames between existing frames. In this paper, we propose a novel frame interpolation technique that uses dominant MV and variance different value in each block. At first, the proposed algorithm performs unidirectional motion estimation using blocking matching algorithm. The new frame is generated by pixel average using compared block variance or by pixel motion compensation using dominant motion vector, whether the motion estimation find the target area or not. Several experiments with the proposed algorithm shows that the proposed algorithm has better image quality than the existing bidirectional frame interpolation algorithm at the rate of about 3dB PSNR and has low complexity comparing to the unidirectional frame interpolation technique.

• Journal of Institute of Control, Robotics and Systems
• /
• v.22 no.9
• /
• pp.745-752
• /
• 2016

We suggest an efficient Simultaneous Localization and 3D Polygon Map Building (SLAM) method with Kinect depth sensor for mobile robots in indoor environments. In this method, Kinect depth data is separated into row planes so that scan line segments are on each row plane. After grouping all scan line segments from all row planes into line groups, a set of 3D Scan polygons are fitted from each line group. A map matching algorithm then figures out pairs of scan polygons and existing map polygons in 3D, and localization is performed to record correct pose of the mobile robot. For 3D map-building, each 3D map polygon is created or updated by merging each matched 3D scan polygon, which considers scan and map edges efficiently. The validity of the proposed 3D SLAM algorithm is revealed via experiments.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.45 no.2
• /
• pp.1-9
• /
• 2008

In this paper, to extract object regions in stereo image, we propose an enhanced algorithm that extracts objects combining both of brightness information and disparity information. The approach that extracts objects using both has been studied by Ping and Chaohui. In their algorithm, the segmentation for an input image is carried out using the brightness, and integration of segmented regions in consideration of disparity information within the previously segmented regions. In the regions where the brightness values between object regions and background regions are similar, however, the segmented regions probably include both of object regions and background regions. It may cause incorrect object extraction in the merging process executed in the unit of the segmented region. To solve this problem, in proposed method, we adopt the merging process which is performed in pixel unit. In addition, we perform the bi-directional stereo matching process to enhance reliability of the disparity information and supplement the disparity information resulted from a single directional matching process. Further searching for disparity is decided by edge information of the input image. The proposed method gives good performance in the object extraction since we find the disparity information that is not extracted in the traditional methods. Finally, we evaluate our method by experiments for the pictures acquired from a real stereoscopic camera.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.38 no.3
• /
• pp.258-268
• /
• 2001

The compression method is necessarily used to send the high quality moving picture that contains a number of data in image processing. In the field of moving picture compression method, the motion estimation algorithm is used to reduce the temporal redundancy. Block matching algorithm to be usually used is distinguished partial search algorithm with full search algorithm. Full search algorithm be used in this paper is the method to compare the reference block with entire block in the search window. It is very efficient and has simple data flow and control circuit. But the bigger the search window, the larger hardware size, because large computational operation is needed. In this paper, we design the full search block matching motion estimator. Using the DCT DC values, we decide luminance. And we apply 3 bit compare-selector using bit plane to I(Intra coded) picture, not using 8 bit luminance signals. Also it is suggested that use the same selective bit for the P(Predicted coded) and B(Bidirectional coded) picture. We compare based full search method with PSNR(Peak Signal to Noise Ratio) for C language modeling. Its condition is the reference block 8$\times$8, the search window 24$\times$24 and 352$\times$288 gray scale standard video images. The result has small difference that we cannot see. And we design the suggested motion estimator that hardware size is proved to reduce 38.3％ for structure I and 30.7％ for structure II. The memory is proved to reduce 31.3％ for structure I and II.

• Journal of Preventive Medicine and Public Health
• /
• v.34 no.3
• /
• pp.253-261
• /
• 2001

Objectives : To evaluate the hypothesis that air pollution could increase emergency room visits for respiratory diseases, and if so, to quantify the strength of association between those. Methods : We compiled daily records of hospital emergency room visits for respiratory diseases in Seoul, from November 1. 1955 to October 31. 1996, by using medical utilization data of unscheduled visits. In addition, air quality and weather data for the same period was collected. And a case-crossover design was applied by adopting conditional logistic regression analysis to determine the relationship between air pollutants and emergency room visits for respiratory diseases. In particular, the control periods were chosen by a bidirectional paired matching technique 7, 14, and 21 days before and after the case periods. Results : Only ozone was associated with the increased number of emergency room visits for respiratory diseases. The relative risk according to a 30ppb increase of ozone concentration (24hr mean, lagged 1day) was 1.91(95% confidence interval = 1.78-2.05). Conclusion : There was a statistically significant association between the ambient ozone and daily emergency room visits for respiratory diseases.