• Journal of Korea Multimedia Society
• /
• v.9 no.10
• /
• pp.1266-1275
• /
• 2006

In this paper, we propose a real-lime face detection and tracking algorithm using AdaBoost(Adaptive Boosting) algorithm. The proposed algorithm consists of two levels such as the face detection and the face tracking. First, the face detection used the eight-wavelet feature models which ate very simple. Each feature model applied to variable size and position, and then create initial feature set. The intial feature set and the training images which were consisted of face images, non-face images used the AdaBoost algorithm. The basic principal of the AdaBoost algorithm is to create final strong classifier joining linearly weak classifiers. In the training of the AdaBoost algorithm, we propose SAT(Summed-Area Table) method. Face tracking becomes accomplished at real-time using the position information and the size information of detected face, and it is extended view region dynamically using the fan-Tilt camera. We are setting to move center of the detected face to center of the Image. The experiment results were amply satisfied with the computational efficiency and the detection rates. In real-time application using Pan-Tilt camera, the detecter runs at about 12 frames per second.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.5
• /
• pp.1431-1445
• /
• 2022

We construct a deep neural network model named ECGResNet. This model can diagnosis diseases based on 12-lead ECG data of eight common cardiovascular diseases with a high accuracy. We chose the 16 Blocks of ResNet50 as the main body of the model and added the Squeeze-and-Excitation module to learn the data information between channels adaptively. We modified the first convolutional layer of ResNet50 which has a convolutional kernel of 7 to a superposition of convolutional kernels of 8 and 16 as our feature extraction method. This way allows the model to focus on the overall trend of the ECG signal while also noticing subtle changes. The model further improves the accuracy of cardiovascular and cerebrovascular disease classification by using a fully connected layer that integrates factors such as gender and age. The ECGResNet model adds Dropout layers to both the residual block and SE module of ResNet50, further avoiding the phenomenon of model overfitting. The model was eventually trained using a five-fold cross-validation and Flooding training method, with an accuracy of 95% on the test set and an F1-score of 0.841.We design a new deep neural network, innovate a multi-scale feature extraction method, and apply the SE module to extract features of ECG data.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.33 no.5
• /
• pp.363-371
• /
• 2015

Feature extraction is one of the main goals in many remote sensing analyses. After high-resolution imagery became more available, it became possible to extract more detailed and specific features. Thus, considerable image segmentation algorithms have been developed, because traditional pixel-based analysis proved insufficient for high-resolution imagery due to its inability to handle the internal variability of complex scenes. However, the individual segmentation method, which simply uses color layers, is limited in its ability to extract various target features with different spectral and shape characteristics. Spectral indices can be used to support effective feature extraction by helping to identify abundant surface materials. This study aims to evaluate a feature extraction method based on a segmentation technique with spectral indices. We tested the extraction of diverse target features-such as buildings, vegetation, water, and shadows from eight band WorldView-2 satellite image using decision tree classification and used the result to draw the appropriate spectral indices for each specific feature extraction. From the results, We identified that spectral band ratios can be applied to distinguish feature classes simply and effectively.

• Journal of Korea Multimedia Society
• /
• v.3 no.3
• /
• pp.241-252
• /
• 2000

In this paper, a translation, rotation and scale invariant system for the recognition of closed 2-D images using the bispectrum of a contour sequence and a weighted fuzzy classifier is derived and compared with the recognition process using one of the competitive neural algorithm, called a LVQ( Loaming Vector Quantization). The bispectrum based on third order cumulants is applied to the contour sequences of an image to extract fifteen feature vectors for each planar image. These bispectral feature vectors, which are invariant to shape translation, rotation and scale transformation, can be used to the represent two-dimensional planar images and are fed into a weighted fuzzy classifier. The experimental processes with eight different shapes of aircraft images are presented to illustrate a relatively high performance of the proposed recognition system.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38A no.2
• /
• pp.174-182
• /
• 2013

Most methods for fusion-based finger vein recognition were to fuse different features or matching scores from more than one trait to improve performance. To overcome the shortcomings of "the curse of dimensionality" and additional running time in feature extraction, in this paper, we propose a finger vein recognition technology based on matching score-level fusion of a single trait. To enhance the quality of finger vein image, the contrast-limited adaptive histogram equalization (CLAHE) method is utilized and it improves the local contrast of normalized image after ROI detection. Gabor features are then extracted from eight channels based on a bank of Gabor filters. Instead of using the features for the recognition directly, we analyze the contributions of Gabor feature from each channel and apply a weighted matching score-level fusion rule to get the final matching score, which will be used for the last recognition. Experimental results demonstrate the CLAHE method is effective to enhance the finger vein image quality and the proposed matching score-level fusion shows better recognition performance.

• Journal of Korea Multimedia Society
• /
• v.11 no.6
• /
• pp.860-868
• /
• 2008

Human detection techniques in outdoor scenes have been studied for a long time to watch suspicious movements or to keep someone from danger. However there are few methods of human detection in overhead or near-field view scenes, while lots of human detection methods in far-field view scenes have been developed. In this paper, a set of five features useful for human detection in overhead view scenes and another set of four useful features in near-field view scenes are suggested. Eight feature-candidates are first extracted by analyzing geometrically varying characteristics of moving objects in samples of video sequences. Then highly contributed features for each view scene to classifying human from other moving objects are selected among them by using a neural network learning technique. Through experiments with hundreds of moving objects, we found that each set of features is very useful for human detection and classification accuracy for overhead view and near-field view scenes was over 90%. The suggested sets of features can be used effectively in a PTZ camera based surveillance system where both the overhead and near-field view scenes appear.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.19 no.1
• /
• pp.192-200
• /
• 2015

Several algorithms have been developed to classify arrhythmia which rely on specific ECG(Electrocardiogram) database. Nevertheless personalized difference of ECG signal exist, performance degradation occurs because of carrying out diagnosis by general classification rule. Most methods require accurate detection of P-QRS-T point, higher computational cost and larger processing time. But it is difficult to detect the P and T wave signal because of person's individual difference. Therefore it is necessary to classify the pattern by analyzing personalized ECG signal and extracting minimal feature. Thus, QRS pattern Analysis of personalized ECG Signal by Q, R, S peak variability is presented in this paper. For this purpose, we detected R wave through the preprocessing method and extract eight feature by amplitude and phase variability. Also, we classified nine pattern in realtime through peak and morphology variability. PVC, PAC, Normal, LBBB, RBBB, Paced beat arrhythmia is evaluated by using 43 record of MIT-BIH arrhythmia database. The achieved scores indicate the average of 93.72% in QRS pattern detection classification.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.2
• /
• pp.1043-1056
• /
• 2017

A styled document is a document that contains diverse decorating functions such as different font, colors, tables and images generally authored in a word processor (e.g., MS-WORD, Open Office). Compared to a plain-text document, a styled document enables a human to easily recognize a logical structure such as section, subsection and contents of a document. However, it is difficult for a computer to recognize the structure if a writer does not explicitly specify a type of an element by using the styling functions of a word processor. It is one of the obstacles to enhance document version management systems because they currently manage the document with a file as a unit, not the document elements as a management unit. This paper proposes a machine learning based approach to analyzing the logical structure of a styled document composing of sections, subsections and contents. We first suggest a feature vector for characterizing document elements from a styled document, composing of eight features such as font size, indentation and period, each of which is a frequently discovered item in a styled document. Then, we trained machine learning classifiers such as Random Forest and Support Vector Machine using the suggested feature vector. The trained classifiers are used to automatically identify logical structure of a styled document. Our experiment obtained 92.78% of precision and 94.02% of recall for analyzing the logical structure of 50 styled documents.

• Archives of design research
• /
• v.20 no.3 s.71
• /
• pp.107-118
• /
• 2007

The aim of this paper is to suggest a design guideline for a teaching assistant robot by finding out images that satisfy the role of the teaching assistant robot, and to search for a body features that show such images. Role images of teaching assistant robots were established from literature review and factor analysis. And eight elements of body features were extracted from human's elements of body feature. Robot external form samples varied according to the body feature was modeled three-dimensionally. Children, who are the main users of teaching assistant robots, valuated the 3D robot samples projected onto wall in real size. The valuation basis was role images of teaching assistant robots, adjectives about age and gender, preference, and appropriateness as teaching assistant robots. The result of valuation was analyzed by analysis of variance, and analysis of correlation. The result revealed the fact that four elements of body feature (the ratio of head length, height, the ratio of breast girth, and waist girth) were related to role images. Among these elements, height and waist girth was more important than the rest, particularly, waist girth had strong relation with all the role images. Also, in order to reveal tender and kind image, the ratio of head length was proved to have to be adjusted according to waist girth. On the basis of these result, a design guideline for a teaching assistant robot was suggested.

• The Journal of the Korea Contents Association
• /
• v.14 no.9
• /
• pp.141-152
• /
• 2014

The user interfaces of electronic-book (e-book) readers in the e-book market are highly diverse, and this has led to major usability issues. In this paper, we analyze user preferences in terms of feature-setting options and the use of touchscreen actions in the six most commonly used e-book readers for smartphnoes. We identify alternatives for these features, which can enhance the usability of e-book readers, based on these user preferences. The survey results for the feature-setting options show that it is desirable to support at least eight background colors, as well as the ability to specify the color of the background icons. Adjusting the screen brightness using a setting bar with the support of an auto-brightness option is desirable, as in using +/- buttons to adjust the font size, as well as approximately 10 font faces. We find that it is desirable to support fade, slide, scroll and curl page-turing options, in addition to a simple non-animated page-turning effect, and that page movement should be accomplished using a scroll bar with the support of the page movement features by entering a page number, and by using the table of contents as well as bookmarks. The survey results on the use of touchscreen options indicated that it is useful to be able to configure the screen layout of an e-book reader by dividing it into three areas: left, right, and center. And then, when users briefly touch the left and right areas, it is ideal to move to the previous and subsequent pages, respectively; and when they briefly touch the center region, it is desirable to display a touch feature menu. We believe that the results of this study may provide guidance in the design of user interfaces for e-book readers.