Earticle

본 논문에서는 엘리베이터 내부 탑승자의 비상상황 감지 시스템을 제안한다. 엘리베이터는 구조적으로 외부와 격리되어 있다. 보통 엘리베이터 내부에서 일어나는 범죄는 성범죄와 폭력사건 등의 중범죄가 대부분이며 이 경우 경비실이나 구조기관에 위험상황을 전달하기에는 문제점이 많다. 본 논문에서 제안하는 시스템은 엘리베이터에서 탑승자의 안전을 보장하기 위하여 위험상황을 감지하고 대응한다. 이를 위해 탑승자에게서 얻을 수 있는 3가지 정보를분석한다. 첫 번째로 카메라로 입력되는 영상, 마이크로 입력되는 탑승자의 성량 및 주변소음, 마지막으로 상황이발생하였을 때 엘리베이터 내부의 충격량을 분석한다. 우선 HOG특징과 차분법, 움직임 벡터를 통해 1인과 다수의움직임에 대해서 분석하고, 성량 크기와 충격센서로 전달되는 충격량을 분석하여 비상상황 유무를 판단한다.

In this paper, we propose an emergency detection system inside the elevator for the passengers. The elevator is structurally isolated from the outside environment. Usually the most serious crimes, such as sexual exploitation and violence may occur inside the elevator. In such situations, there are many ways that message about such situations can be passed to the security office or agency. The proposed system detects a dangerous situation, in order to ensure the safety and the support for the passenger from the elevator. For this purpose, we analyze the three pieces of information that can be obtained from the occupant. First, input image using the camera, next input voice by a microphone and around the occupant input noise, and finally, the analysis of the impulse inside the lift where this situation occurs. First, we use the histogram of gradient (HOG) to identify whether the motion is of the single person or of the group of people with the motion vector. Next, the impulse generated by voice dB and impact sensor determines whether there is an emergency situation or not.

This paper describes the algorithm that lowers the dimension, maintains the gesture recognition and significantly reduces the eigenspace configuration time by combining the higher correlation feature information and Principle Component Analysis. Since the suggested method doesn't require a lot of computation than the method using existing geometric information or stereo image, the fact that it is very suitable for building the real-time system has been proved through the experiment. In addition, since the existing point to point method which is a simple distance calculation has many errors, in this paper to improve recognition rate the recognition error could be reduced by using several successive input images as a unit of recognition with K-Nearest Neighbor which is the improved Class to Class method.

본 논문은 음악 정보검색에 사용되는 효과적인 템포 특징 추출방식을 제안한다. 제안된 템포 정보는 협소 밴드상의 일시적인 변조 성분에 의해 형성된다. 이러한 변조 성분은 시간 축 상의 음악 신호로부터 스펙트럼을 구한 후, 각 스펙트럼 성분에 대한 주파수 영역 분석을 통해 획득된 변조 스펙트럼으로 구성된다. 실제 구현에 있어서는 MP3 음악파일로부터 부분 디코딩에 의해 출력된 변형된 이산 코사인 변환 계수에 퓨리에 변환을 취하여 변조스펙트럼을 구하였다. 획득된 변조 스펙트럼의 진폭으로부터 고속으로 추출된 음악 템포 특징값은 다양한 음악 정보 검색에 적용되었다. 음악 무드 및 장르 분류에서는 로그 변조 주파수 계수를 적용하여 분류 성능을 개선시켰으며, 적응 변조 스펙트럼에서 유도된 비트 벡터는 오디오 핑거프린팅에 적용되어 잡음환경 하에서도 검색 성능을 크게 향상시켰다.

This paper proposes an effective tempo feature extraction method for music information retrieval. The tempo information is modeled by the narrow-band temporal modulation components, which are decomposed into a modulation spectrum via joint frequency analysis. In implementation, the tempo feature is directly extracted from the modified discrete cosine transform coefficients, which is the output of partial MP3(MPEG 1 Layer 3) decoder. Then, different features are extracted from the amplitudes of modulation spectrum and applied to different music information retrieval tasks. The logarithmic scale modulation frequency coefficients are employed in automatic music emotion classification and music genre classification. The classification precision in both systems is improved significantly. The bit vectors derived from adaptive modulation spectrum is used in audio fingerprinting task That is proved to be able to achieve high robustness in this application. The experimental results in these tasks validate the effectiveness of the proposed tempo feature.

얼굴의 정보가 전체적으로 처리되는가 혹은 국소적으로 처리되는가 하는 것은 얼굴인식과 관련된 중요한 논쟁 중의 하나이다. 최근에는 전체적 처리와 국소적 처리가 동시에 이루어지며, 각기 다른 경로를 통해 이루어진다는 주장이 제기된 바 있다. 본 연구에서는 얼굴 정보에 대한 이러한 이중경로처리설을 확인하기 위하여, 형태정보처리 경로의 시각 세포들이 최적으로 반응하는 자극속성을 반영한 방사주파수 합성얼굴을 이용하여 얼굴의 윤곽과 국소정보에 대한 선택적 뇌 활성화의 차이를 fMRI측정을 통하여 살펴보았다. 실험 1에서는 얼굴의 윤곽과 국소정보 모두에 대하여 FFA 영역에서의 활성화를 관찰하였다. 주목할 점은, 국소정보에 비하여 윤곽정보에 대한 FFA에서의 활성화가 더 강하게 관찰된 것이다. 실험 2에서는 얼굴의 윤곽정보에 대한 전체적 처리가 인출과정보다는 부호화과정에 관여함을 관찰하였고, 국소정보의 부호화 조건에서는 전전두엽(BA 10)과 대상회(BA 32)에서의 강한 활성화를 관찰하였다. 이러한 결과들은, 얼굴인식이 이중경로를 통해 이루어진다는 연구들을 지지하는 증거로 해석되며, 얼굴의 윤곽정보는 상향처리를 통해, 국소정보는 하향처리를 통해 인식되는 것임을 시사한다.

One of the controversial issues on face recognition is whether faces are recognized as undifferentiated wholes or in terms of their constituent parts, namely, global vs local information processing for face recognition. However, it has been recently proposed that global coding and local processing constitute dual routes to face recognition. To investigate this dual routs processing hypothesis, we directly examined the selective activation of human brain areas with fMRI measurements to the synthetic face stimuli composed of radial frequency components, which had an advantage to easily separate the global contour and local basic feature information of face. In experiment 1, we found that FFA was activated to face contour and feature information. More importantly, it was observed that the strength of activation to contour information was higher than to feature information in FFA. In experiment 2, we also found that the global processing of face contour information was involved mainly in encoding processing, not in retrieval processing. Strong activations of the prefrontal region (BA10) and the cingulate gyrus (BA 32) were observed during encoding processing of face feature information. These results altogether add to our understanding of the characteristics of dual routes processing in face recognition. In addition, these results suggest that the contour information of face stimulus is processed through a bottom-up processing whereas the feature information is processed through a top-down processing.

본논문에서는Different 기반이상탐지를위한재구성 모델로Dense Prediction Transformer, U-Net, ResU-Net 모델을 선정하고, 해당 딥러닝 모델의 Decoder에 전달되는 특징 정보량을 조절함으로써 변동되는 재구성 결과에 대한 분석을 진행 하였다. 제시한각모델은MVTec 데이터셋내Bottle 클래스의정상데이터만을학습한후결함이존재하는비정상데이터 를 재구성하며, 세밀한 표현이 가능한 선정 모델들의 특징 정보량을 직접 조절함으로써 단순한 Encoder-Decoder 구조의 방 식의한계인세밀한특징표현이어려운문제를해소하고, 높은해상도의재구성이가능하게한다. 최종적으로각모델의결 과에 대한 정성적 평가를 진행하여 입력되는 비정상 데이터와 재구성되는 데이터의 차이를 확인하였으며, 비교 모델 중 Dense Prediction Transformer 모델이Different 기반이상탐지기법의기반모델로가장적합한모델임을확인하였다.

ICT 기술의 발전과 스마트폰 보급의 활성화로 온라인 마켓을 통해 다양한 제품을 구매하는 구매 서비스가 활성화되 고 있다. 특히 보관 및 배송 기술의 발전으로 인하여 보관기관이 짧은 식자재의 경우도 온라인을 통해 구매가 가능함으로써 오프라인 판매만을 수행하는 상가의 경우 매출이 감소되고 있는 추세이다. 따라서 본 논문에서는 기존 오프라인 판매만 수행이 가능한 소규모 상가에서 전문적인 판매 지식 및 판매망이 없어도 구매 서비스를 통해 광고 효과 및 주문, 배달이 가능한 통합 솔루션을 제안한다. 제안하는 시스템은 사용자가 원하는 제품에 대한 이미지 검색을 통해 효율적으로 제품에 대한 정보를 카테고리별로 볼 수 있으며, 이로 인해 등록된 제품 판매처가 추가적인 광고가 없어도 효율적으로 판매가 가능하다는 장점이 있다.

With the development of ICT technology and the promotion of smartphone penetration, purchasing services that purchase various products through online market are being activated. In particular, due to advances in storage and delivery technology, sales of short food materials can be purchased online. Therefore, in this paper, we propose an integrated solution that enables advertisement effect, ordering and delivery through a purchase service even if there is no sales knowledge and sales network in a small shopping mall where only offline sales can be performed. The proposed system is able to efficiently view the product information by category through image search for the product that the user desires, so that the seller of the registered product can efficiently sell without any additional advertisement.

This paper, using the speech thinking of protocol analysis, tried to improve the reliability by quantified using a multivariate analysis to process that categorizes the results obtained from interviews conducted interviews. Japanese, “happy”, “fear, sadness,” “surprise, anger” are plotted in a short distance. “Fear, sadness” is not distinguish, for example, between large and small and the voice of the low of the voice as a clue, “surprise, anger,” it was found to be distinguished, such as How to put out of breath and the strength of the voice as a clue. Koreans, “fear‐sad” is have been plotted in a short distance, “happy, anger, surprise,” are plotted at a short distance. For Korean, it said to have been largely based on the overall voice intensity determination. Chinese “happy”, “fear”, “sad”, “anger, surprise” has been plotted at a short distance. Chinese plotted is different from speech feature position is different plot position by country of Japanese and Korean than experiment collaborators.

In view of the greater changes of posture, illumination, expression and scene in reality environment have a strong impact on wild face recognition algorithm to identify performance problem, and puts forward a kind of linear discriminant analysis side information (SILD) algorithm on hyperplane fusion of learning prototype. First of all, using support vector machine (SVM) to weak tag of data-concentrated sample is expressed as the middle-level characteristics of prototype hyperplane, using a learning combination coefficient to select sparse support vector set from untagged conventional data set; then, under the constraints of the combination sparse coefficient of SVM model, by using Fisher linear discriminant criterion to maximize discriminant ability of untagged data set, and using the iterative optimization algorithm to solve the objective function; in the end, using SILD for feature extraction, cosine similarity measure to complete the final face recognition. In two general face data sets of wild face recognition (LFW) and YouTube, it makes comparison of PHL+SILD method and low-level features + SILD method on some characteristics, such as strength, LBP, Gabor feature and Block Gabor feature, average accuracy, area under the curve (AUC) and entire error rate (EER). The validity and reliability of the proposed algorithm is verified by the experiments.

Semantic role labeling is typically used to resolve a problem from the perspective of classification by using a corpus. Semantic Role Labeling determines an adequate predicate-argument relation. It can be used to improve performance in various areas of natural language processing. In this paper, automatic semantic role labeling using 10,000 sentences in a semantic role tagged corpus constructed from a Korean syntax tagged corpus was conducted. In Korean, affix, such as josa and eomi, is a very important role in syntactic parsing and semantic role labeling. Semantic role labeling was achieved in this study by improving particle and word ending information, which were insufficiently addressed in previous studies on semantic role labeling, and creating new features. When features based on the affix information created in this study were added to the basic features used in previous studies on semantic role labeling of languages, an F1 score of approximately 80.83% was obtained.

CHI statistical text feature selection method based on information entropy optimization is presented in this paper. In the text categorization process of feature selection, considering the results of effect of the distribution within categories and among categories, we introduce the frequency of features information entropy among categories, the information entropy within categories, information within category to optimize the CHI statistical methods. The experimental results show that the classification accuracy of the optimized CHI method is significantly higher than that the traditional CHI statistical methods.

With the rapid development of the Internet, the application of data mining in the Internet is becoming more and more extensive. However, the data source’s complex feature redundancy leads that data mining process becomes very inefficient and complex. So feature selection research is essential to make data mining more efficient and simple. In this paper, we propose a new way to measure the correlation degree of internal features of dataset which is a mutation of mutual information. Additionally we also introduce Hoeffding inequality as constraint of constructing algorithm. During the experiments, we use C4.5 classification algorithm as test algorithm and compare HSF with BIF(feature selection algorithm based on mutual information). Experiments results show that HSF performances better than BIF[1] in TP and FP rate, what’s more the feature subset obtained by HSF can significantly improve the TP, FP and memory usage of C4.5 classification algorithm.

DNA microarray technique can detect tens of thousands of genes activity in cells and has been widely used in clinical diagnosis. However, microarray data has characteristics of high dimension and small samples, moreover many irrelevant and redundant genes also decrease performance of classification algorithm .Mutual information is very effective method and has widely been used in feature gene selection, but it cannot directly deal with continuous features. Therefore, this paper proposes a novel feature gene selection method to resolve this problem. Firstly, a lot of irrelevant genes are eliminated from original data by using reliefF algorithm , and the candidate subset of genes is obtained; Secondly, a algorithm based on neighborhood mutual information and forward greedy search strategy which deals with directly continuous features is proposed to select feature genes in above genes subset. Here, because radius of neighborhood greatly affects reduction performance, differential evolution algorithm is applied to optimize radius before reduction. The simulation results on six benchmark microarray datasets show that our method can obtain higher classification accuracy using as few genes as possible, especially neighborhood mutual information can directly continuous features. Feature genes selected has an important meaning for understanding microarray data and finding pathogenic genes of cancer. It is an effective and efficient method for feature genes selection.

DNA microarray technique can detect tens of thousands of genes activity in cells and has been widely used in clinical diagnosis. However, microarray data has the characteristics of high dimension and small samples, moreover many irrelevant and redundant genes also decrease performance of classification algorithm. Feature gene selection is an effective method to solve this problem. This paper proposes a hybrid feature gene selection method. Firstly, a lot of irrelevant genes from original data were eliminated by using reliefF algorithm, and the candidate feature genes subset is obtained; Secondly, Fuzzy neighborhood rough set with information entropy which deals directly with continuous data is proposed to reduce redundant genes among genes subset above. Here, differential evolution algorithm is used to optimize radius before reduction by using fuzzy neighborhood rough set, because radius of neighborhood greatly affects reduction performance. The simulation results on six microarray datasets indicate that our method can obtain higher classification accuracy by using as few genes as possible, especially feature genes selected are important for understanding microarray data and identifying the pathogenic genes. The results demonstrated that this method is effective and efficient for feature genes selection.

특징 기반 영상 인식 방법은 객체의 특징을 이용하므로 템플릿 정합에 비해 고속으로 수행될 수 있다. 불변 특징 기반의 파노라마 생성은 영상 인식의 한 응용으로서, 두 영상 간의 특징점 정합에 많은 처리 시간이 필요하다. 본 논문에서는 특징 강도 정보를 이용하여 특징점 정합 속도를 향상시키는 방법을 제안한다. SURF 알고리즘으로 특 징점들을 추출한 후, 특징 강도 정보를 계산하여 강한 특징점들을 선택하여 특징 정합에 사용한다. 특징 강도가 강한 특징점들은 그렇지 않은 특징점들 보다 더 의미 있다고 볼 수 있다. 실험에서 320×240 크기의 칼라 영상에 대해 제 안한 방법은 특징 강도 정보를 사용하지 않았을 때보다 40% 이상 처리 속도의 향상을 보였다.

A feature-based image recognition method, using features of an object, can be performed faster than a template matching technique. Invariant feature-based panoramic image generation, an application of image recognition, requires large amount of time to match features between two images. This paper proposes a speed-up method of feature matching using feature strength information. Our algorithm extracts features in images, computes their feature strength information, and selects strong features points which are used to match the selected features. The strong features can be referred to as meaningful ones than the weak features. In the experiments, it was shown that our method speeded up over 40% of processing time than the technique without using feature strength information.

현재 RCS-e(Rich Communication Service-e)는 복잡한 프로토콜과 닫힌 외부 인터페이스로 인해 일반 개발 자들이 쉽게 접근할 수 없는 실정이다. RCS-e 서비스를 보다 활성화 시키기 위해서는 일반 개발자들이 쉽게 자신의 앱에서 RCS-e 서비스를 이용할 수 있도록 프로토콜을 단순화하고 외부에서 쉽게 접근할 수 있도록 API(Application Program Interface)화 해야 한다. 본 논문은 RCS-e의 여러 기능 중 프레즌스 기능을 웹 인터페이스로 이용할 수 있 게 해주는 오픈 API 프레임워크를 제안한다. 이를 위해 각 노드들 간의 연관 관계를 나타낸 시스템 구성도를 설계하였 다. 프레즌스 기능 제공을 위한 오픈 API 프레임워크와 기존 노드들 간의 메시지 흐름도를 정의하였다. 그리고 오픈 API 프레임워크에서 웹 기반의 요청을 어떻게 RCS-e에서 사용되는 요청으로 변환하는지 예시를 들어 설명하였다. 성 능 평가에서는 오픈 API 프레임워크를 도입해도 기존 인프라의 성능에 영향을 끼치지 않는다는 것을 증명하였다.

Web developers have had difficulties in using Rich Communication Service-e(RCS-e) on their applications because of complicated protocols and closed interfaces. In order to vitalize the use of RCS-e, we need a RCS Application Program Interface(API) which has simple protocols and can be accessed easily. This paper presents the web-based Open API Framework for the RCS-e presence feature. A system architecture for the framework is defined. Call flows for the presence feature between the framework and other nodes are defined. Also, one of the call flows is illustrated to explain how to convert web-based requests to RCS-e requests. Finally, performance evaluation proves that the framework does not add any loads to the existing network infrastructure.

As technology related to digital imaging equipment is developed and generalized, digital imaging system is used for various purposes in fields of society. The object tracking technology from digital image data in real time is one of the core technologies required in various fields such as security system and robot system. Among the existing object tracking technologies, cam shift technology is a technique of tracking an object using color information of an object. Recently, digital image data using infrared camera functions are widely used due to various demands of digital image equipment. However, the existing cam shift method can not track objects in image data without color information. Our proposed tracking algorithm tracks the object by analyzing the color if valid color information exists in the digital image data, otherwise it generates the lightness feature information and tracks the object through it. The brightness feature information is generated from the ratio information of the width and the height of the area divided by the brightness. Experimental results shows that our tracking algorithm can track objects in real time not only in general image data including color information but also in image data captured by an infrared camera.

The statistical analysis of the feature extraction and the neural networks are proposed to recognize a human face. In the preprocessing step, the normalized skin color map with Gaussian functions is employed to extract the region efface candidate. The feature information in the region of face candidate is used to detect a face region. In the recognition step, as a tested, the 360 images of 30 persons are trained by the backpropagation algorithm. The images of each person are obtained from the various direction, pose, and facial expression, Input variables of the neural networks are the feature information that comes from the eigenface spaces. The simulation results of 30 persons show that the proposed method yields high recognition rates.

Battiti's mutual information feature selector (MIFS) and its variant algorithms are used for many classification applications. Since they ignore feature synergy, MIFS and its variants may cause a big bias when features are combined to cooperate together. Besides, MIFS and its variants estimate feature redundancy regardless of the corresponding classification task. In this paper, we propose an automated greedy feature selection algorithm called conditional mutual information-based feature selection (CMIFS). Based on the link between interaction information and conditional mutual information, CMIFS takes account of both redundancy and synergy interactions of features and identifies discriminative features. In addition, CMIFS combines feature redundancy evaluation with classification tasks. It can decrease the probability of mistaking important features as redundant features in searching process. The experimental results show that CMIFS can achieve higher best-classification-accuracy than MIFS and its variants, with the same or less (nearly 50%) number of features.

Cultural heritage is a special object which has the historical identity. Therefore cultural heritage must be protected before being damaged. The most important thing to know about cultural heritage is that once cultural heritage is destroyed, restoration does not return cultural property to its original condition. Cultural heritage information management system becomes more important as archival resources are necessary for restoration and reproduction of cultural heritage. In 2017 Cultural Heritage Administration integrated the system for effective management and in 2018 tried to integrate the different monitoring forms used by 21 different organizations. In this research, the feature to the integrated information management system by grasping the characteristics of the cultural heritage ordinary monitoring system in the development stage will be analyzed. at the points of 1. monitoring and repair cycle ; 2. professionalism ; 3. variety of targets.

Vocal effort detection is important for both robust speech recognition and speaker recognition. In this paper, the spectral information entropy feature which contains more salient information regarding the vocal effort level is firstly proposed. Then, the model fusion method based on complementary model is presented to recognize vocal effort level. Experiments are conducted on isolated words test set, and the results show the spectral information entropy has the best performance among the three kinds of features. Meanwhile, the recognition accuracy of all vocal effort levels reaches 81.6%. Thus, potential of the proposed method is demonstrated.