• Journal of Internet Computing and Services
• /
• v.13 no.4
• /
• pp.1-10
• /
• 2012

In this paper, an optimal feature selection method considering sematic of features, which is preprocess of document classification is proposed. The feature selection is very important part on classification, which is composed of removing redundant features and selecting essential features. LSA (Latent Semantic Analysis) for considering meaning of the features is adopted. However, a supervised LSA which is suitable method for classification problems is used because the basic LSA is not specialized for feature selection. We also apply GA (Genetic Algorithm) to the features, which are obtained from supervised LSA to select better feature subset. Finally, we project documents onto new selected feature subset and classify them using specific classifier, SVM (Support Vector Machine). It is expected to get high performance and efficiency of classification by selecting optimal feature subset using the proposed hybrid method of supervised LSA and GA. Its efficiency is proved through experiments using internet news classification with low features.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.8
• /
• pp.3984-4005
• /
• 2017

Image recognition has become an increasingly important topic for its wide application. It is highly challenging when facing to large-scale database with large variance. The recognition systems rely on a key component, i.e. the low-level feature or the learned mid-level feature. The recognition performance can be potentially improved if the data distribution information is exploited using a more sophisticated way, which usually a function over hidden variable, model parameter and observed data. These methods are called generative score space. In this paper, we propose a discriminative extension for the existing generative score space methods, which exploits class label when deriving score functions for image recognition task. Specifically, we first extend the regular generative models to class conditional models over both observed variable and class label. Then, we derive the mid-level feature mapping from the extended models. At last, the derived feature mapping is embedded into a discriminative classifier for image recognition. The advantages of our proposed approach are two folds. First, the resulted methods take simple and intuitive forms which are weighted versions of existing methods, benefitting from the Bayesian inference of class label. Second, the probabilistic generative modeling allows us to exploit hidden information and is well adapt to data distribution. To validate the effectiveness of the proposed method, we cooperate our discriminative extension with three generative models for image recognition task. The experimental results validate the effectiveness of our proposed approach.

• Journal of the Korean Society for Library and Information Science
• /
• v.43 no.3
• /
• pp.163-180
• /
• 2009

In this paper, we describe a developmental system for establishing personal information tendency based on user queries. For each query, the system classified it based on the category information using a kNN classifier. As category information, we used DDC field which is already assigned to each record in the database. The system accumulates category information for all user queries and the user's personalized feature for the target database. We then developed a personalized retrieval system reflecting the personalized feature to produce search result. Our system re-ranks the result documents by adding more weights to the documents for which categories match with the user's personalized feature. By using user's tendency information, the ambiguity problem of the word could be solved. In this paper, we conducted experiments for personalized search and word sense disambiguation (WSD) on a collection of Korean journal articles of science and technology arena. Our experimental result and user's evaluation show that the performance of the personalized search system and WSD is proved to be useful for actual field services.

• Journal of Korea Multimedia Society
• /
• v.3 no.2
• /
• pp.112-120
• /
• 2000

In this paper, we implement a real-time textile dimension inspection system. It can detect various types of real defects which determine the quality of fabric product, defect positions of textile, classify the distortion angel of moving textile and the density. In the implemented system, we measure the density of textile using zone-crossing method with optical lens to solve the noise and real-time problems. And we compensate distortion angel of textile with the classification of distortion types using gaussian gradient and mean gradient features. And also, it detecs real defects of textile and its positions using gray level co-occurrence matrix features. The implemented texile demension inspection systemcan inspect textile dimensions such as density, distortion angle, defect of textile and defect position at real-time. In the implemented proposed texitile dimension inspection system, It is possible to calculate density and detect default of textile at real-time dimension inspection system, it is possible to calculate density and detect default of textile at textile states throughout at all the significant working process such as dyeing, manufacturing, and other texitle processing.

• Proceedings of the Korean Information Science Society Conference
• /
• 2004.04b
• /
• pp.541-543
• /
• 2004

능동적 학습(active learning)은 제한된 시간과 인력으로 가능한 정확도가 높은 분류기(classifier)를 생성하기 위하여, 훈련집합에 추가할 예제 즉 문의예제(query example)의 선정과 확장된 훈련집합으로 다시 학습하는 과정을 반복하여 수행한다. 능동적 학습의 핵심은 사용자에게 카테고리(category) 부여를 요청할 문의예제를 선정하는 과정에 있다. 효과적인 문의예제를 선정하기 위하여 다양한 방안들이 제안되었으나, 이들은 매 문의단계마다 하나의 문의예제를 선정하는 경우에 가장 적합하도록 고안되었다. 능동적 학습이 복수의 예제를 사용자에게 문의할 수 있다면, 사용자는 문의예제들을 서로 비교해 가면서 작업할 수 있으므로 카테고리 부여작업을 보다 빠르고 정확하게 수행할 수 있을 것이다. 또한 충분한 인력을 보유한 상황에서는, 카테고리 부여작업을 병렬로 처리할 수 있어 전반적인 학습시간의 단축에 큰 도움이 될 것이다. 하지만, 각 예제의 문의예제로써의 적합 정도를 추정하면 유사한 예제들은 서로 비슷한 수준으로 평가되므로, 기존의 방안들을 복수의 문의예제 선정작업에 그대로 적용할 경우, 유사한 예제들이 문의예제로 동시에 선정되어 능동적 학습의 효율이 저하되는 현상이 나타날 수 있다. 본 논문에서는 특정 예제를 문의예제로 선정하면 이와 일정 수준이상 유사한 예제들은 해당 예제와 함께 문의예제로 선정하지 않음으로써, 이러한 문제점을 극복할 수 있는 방안을 제안한다. 제안한 방안을 문서분류 문제에 적용해 본 결과 기존 문의예제 선정방안으로 복수 문의예제를 선정할 때 발생할 수 있는 문제점을 상당히 완화시킬 있을 뿐 아니라, 복수의 문의예제를 선정하더라도 각 문의 단계마다 하나의 예제를 선정하는 경우에 비해 큰 성능의 저하가 없음을 실험적으로 확인하였다./$m\ell$로 나타났다.TEX>${HCO_3}^-$ 이온의 탈착은 서서히 진행되었다. R＆D investment increases are directly not liked to R＆D productivities because of delays and side effects during transition periods between different stages of technology development. Thus, It is necessary to develope strategies in order to enhance efficiency of technological development process by perceiving the switching pattern. 기여할 수 있을 것으로 기대된다. 것이다.'ity, and warm water discharges from a power plant, etc.h to the way to dispose heavy water adsorbent. Through this we could reduce solid waste products and the expense of permanent disposal of radioactive waste products and also we could contribute nuclear power plant run safely. According to the result we could keep the best condition of radiation safety super vision and we could help people believe in safety with Radioactivity wastes control for harmony with Environ

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.10a
• /
• pp.1019-1022
• /
• 2019

최근의 영상 처리 분야는 딥러닝 기법들의 성능이 입증됨에 따라 다양한 분야에서 이와 같은 기법들을 활용해 영상에 대한 분류, 분석, 검출 등을 수행하려는 시도가 활발하다. 그중에서도 의료 진단 보조 역할을 할 수 있는 의료 영상 분석 소프트웨어에 대한 기대가 증가하고 있는데, 본 연구에서는 캡슐내시경 영상에 주목하였다. 캡슐내시경은 주로 소장 촬영을 목표로 하며 식도부터 대장까지 약 8~10시간 동안 촬영된다. 이로 인해 CT, MR, X-ray와 같은 다른 의료 영상과 다르게 하나의 데이터 셋이 10~15만 장의 이미지를 갖는다. 일반적으로 캡슐내시경 영상을 판독하는 순서는 위장관 교차점(Z-Line, 유문판, 회맹판)을 기준으로 위장관 랜드마크(식도, 위, 소장, 대장)를 구분한 뒤, 각 랜드마크 별로 병변 정보를 찾아내는 방식이다. 그러나 워낙 방대한 영상 데이터를 가지기 때문에 의사 혹은 의료 전문가가 영상을 판독하는데 많은 시간과 노력이 소모되고 있다. 본 논문의 목적은 캡슐내시경 영상의 판독에서 모든 환자에 대해 공통으로 수행되고, 판독하는 데 많은 시간을 차지하는 위장관 랜드마크를 찾는 것에 있다. 이를 위해, 위장관 랜드마크를 식별할 수 있는 CNN 학습 모델을 설계하였으며, 더욱 효과적인 학습을 위해 전처리 과정으로 학습에 방해가 되는 학습 노이즈 영상들을 제거하고 위장관 랜드마크 별 특징 분석을 진행하였다. 총 8명의 환자 데이터를 가지고 학습된 모델에 대해 평가 및 검증을 진행하였는데, 무작위로 환자 데이터를 샘플링하여 학습한 모델을 평가한 결과, 평균 정확도가 95% 가 확인되었으며 개별 환자별로 교차 검증 방식을 진행한 결과 평균 정확도 67% 가 확인되었다.

• KIPS Transactions on Computer and Communication Systems
• /
• v.1 no.1
• /
• pp.11-20
• /
• 2012

Stream data shows a sequence of values changing continuously over time. Due to the nature of stream data, its trend is continuously changing according to various time intervals. Therefore the prediction of stream data must be carried out simultaneously with respect to multiple intervals, i.e. Continuous Multiple Prediction(CMP). In this paper, we propose a Continuous Integrated Hierarchical Temporal Memory (CIHTM) network for CMP based on the Hierarchical Temporal Memory (HTM) model which is a neocortex leraning algorithm. To develop the CIHTM network, we created three kinds of new modules: Shift Vector Senor, Spatio-Temporal Classifier and Multiple Integrator. And also we developed learning and inferencing algorithm of CIHTM network.

• KIPS Transactions on Software and Data Engineering
• /
• v.2 no.3
• /
• pp.209-218
• /
• 2013

Early researches in human action recognition have focused on tracking and classifying articulated body motions. Such methods required accurate segmentation of body parts, which is a sticky task, particularly under realistic imaging conditions. Recent trends of work have become popular towards the use of more and low-level appearance features such as spatio-temporal interest points. Given the great progress in pose estimation over the past few years, redefined views about pose-based approach are needed. This paper addresses the issues of whether it is sufficient to train a classifier only on low-level appearance features in appearance approach and proposes effective pose-based approach with pose estimation for emotional action recognition. In order for these questions to be solved, we compare the performance of pose-based, appearance-based and its combination-based features respectively with respect to scenario of various emotional action recognition. The experiment results show that pose-based features outperform low-level appearance-based approach of features, even when heavily spoiled by noise, suggesting that pose-based approach with pose estimation is beneficial for the emotional action recognition.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.22 no.6
• /
• pp.748-753
• /
• 2012

In this paper, Face recognition algorithm is designed based on optimized pRBFNNs pattern classifier using three-dimensional scanner. Generally two-dimensional image-based face recognition system enables us to extract the facial features using gray-level of images. The environmental variation parameters such as natural sunlight, artificial light and face pose lead to the deterioration of the performance of the system. In this paper, the proposed face recognition algorithm is designed by using three-dimensional scanner to overcome the drawback of two-dimensional face recognition system. First face shape is scanned using three-dimensional scanner and then the pose of scanned face is converted to front image through pose compensation process. Secondly, data with face depth is extracted using point signature method. Finally, the recognition performance is confirmed by using the optimized pRBFNNs for solving high-dimensional pattern recognition problems.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.3
• /
• pp.143-150
• /
• 2016

In this paper, we propose Korean monophthong recognition method optimizing muscle mixing based on facial surface EMG signals. We observed that EMG signal patterns and muscle activity may vary according to Korean monophthong pronunciation. We use RMS, VAR, MMAV1, MMAV2 which were shown high recognition accuracy in previous study and Cepstral Coefficients as feature extraction algorithm. And we classify Korean monophthong by QDA(Quadratic Discriminant Analysis) and HMM(Hidden Markov Model). Muscle mixing optimized using input data in training phase, optimized result is applied in recognition phase. Then New data are input, finally Korean monophthong are recognized. Experimental results show that the average recognition accuracy is 85.7% in QDA, 75.1% in HMM.