• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.64 no.7
• /
• pp.1064-1073
• /
• 2015

In this study, we introduce the pedestrian detection system by using the feature of HOG-PCA and RBFNNs pattern classifier. HOG(Histogram of Oriented Gradient) feature is extracted from input image to identify and recognize a object. And a dimension is reduced for improving performance as well as processing speed by using PCA which is a typical dimensional reduction algorithm. So, the feature of HOG-PCA through the dimensional reduction by using PCA leads to the improvement of the detection rate. FCM clustering algorithm is used instead of gaussian function to apply the characteristic of input data as well and connection weight is used by polynomial expression such as constant, linear, quadratic and modified quadratic. Finally, INRIA person database known as one of the benchmark dataset used for pedestrian detection is applied for the performance evaluation of the proposed classifier. The experimental result of the proposed classifier are compared with those studied by Dalal.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.6
• /
• pp.464-468
• /
• 2009

In this paper we propose an image-based approach, which is different from the traditional flow cytometric method to detect shape of apoptosis cells. This method can overcome the defects of cytometry and give precise recognition of apoptosis cells. In this work K-means clustering was used to do the rough segmentation and an active contour model, called 'snake' was used to do the precise edge detection. And then some features were extracted including physical feature, shape descriptor and texture features of the apoptosis cells. Finally a Mahalanobis distance classifier classifies the segmentation images as apoptosis and non-apoptosis cell.

• Korean journal of food and cookery science
• /
• v.28 no.6
• /
• pp.721-729
• /
• 2012

This paper aimed to investigate the influence of consumer lifestyle on consumer selection of bakery cafe attributes. Data were collected through a self-administered questionnaire by 403 random consumers between the ages 20s and 30s in several bakery cafes in Seoul and Gyonggi area. Different methods of statistical analysis had been used such as frequency analysis, factor analysis, k-means clustering analysis, cross tabulation, one way ANOVA and Duncan's multiple range test with SPSS for Window 13.0 package. First, when analyzing the 16 questions of comsumer lifestyles, four factors were extracted: 'dining out-oriented factor', 'achievement-oriented factor', 'brand-oriented factor', and 'health-oriented factor'. Second, the respondents were divided into three groups by k-means cluster analysis: no interest group, dining-out & value oriented group, and health-brand oriented group. Third, consumer's bakery cafe attributes were categorized into five factors including 'food', 'convenience and image', 'store promotion', 'positive dining experience', and 'menu & merchandises'. Finally when analyzing the differences in the selection of bakery cafe attributes according to consumer's lifestyles, it showed a significant differences.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2012.07a
• /
• pp.294-297
• /
• 2012

본 논문에서는 컬러 영상의 검색을 위하여 영상을 색상 정보에 기반한 pixel layer (cluster)의 집합체로 모델링하고, 두 layer 간의 유사도를 각 layer 를 이루는 pixel 들의 색상 분포에 따른 공간적 분포를 이용하여 측정하는 기법을 제안한다. 먼저 pixel layering 단계에서는 HSV 색 공간에서 mean-shift clustering 알고리즘을 통해 초기 layer 들을 얻고, 비슷한 색상의 layer 들은 합쳐 영상의 soft segmentation 과 유사한 결과를 얻는다. 비교할 두 영상에서 pixel layering 을 한 후, 각 layer 를 이진화된 공간분포 지도로 형성하고 그 차이를 비교함으로써 유사도를 측정한다. 이 때, 사용하는 가중치로서 HSV 색 공간 분포의 비슷한 정도를 정의하는데, 이는 HSV 색 공간을 XYZ 의 3 차원 좌표로 설정하고, overlap 되는 pixel 수로 정의하였다. 본 논문에서 제안한 pixel layer 들 간의 색상 공간 분포에 따른 공간적 분포를 이용한 영상 검색 기법은 MPEG-7 에서 정의한 대표색상 기반의 영상 검색보다 우수한 성능을 보여주었다.

• Smart Media Journal
• /
• v.9 no.1
• /
• pp.23-29
• /
• 2020

This paper presents an approach for dynamic hand gesture recognition by using algorithm based on 3D Convolutional Neural Network (3D_CNN), which is later extended to 3D Residual Networks (3D_ResNet), and the neural network based key frame selection. Typically, 3D deep neural network is used to classify gestures from the input of image frames, randomly sampled from a video data. In this work, to improve the classification performance, we employ key frames which represent the overall video, as the input of the classification network. The key frames are extracted by SegNet instead of conventional clustering algorithms for video summarization (VSUMM) which require heavy computation. By using a deep neural network, key frame selection can be performed in a real-time system. Experiments are conducted using 3D convolutional kernels such as 3D_CNN, Inflated 3D_CNN (I3D) and 3D_ResNet for gesture classification. Our algorithm achieved up to 97.8% of classification accuracy on the Cambridge gesture dataset. The experimental results show that the proposed approach is efficient and outperforms existing methods.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2013.10a
• /
• pp.103-107
• /
• 2013

In this paper, A real-time lane tracking algorithm is proposed for lane departure warning system. To eliminate perspective effect, input image is converted into Bird's View by inverse perspective mapping. Next, suitable features are extracted for lane detection. Lane feature that correspond to area of interest and RANSAC are used to detect lane candidates. And driving lane is decided by clustering of lane candidates. Finally, detected lane is tracked using the Kalman filter. Experimental results show that the proposed algorithm can be processed within 30ms and its detection rate is approximately 90% on the highway in a variety of environments such as day and night.

• Journal of Information Processing Systems
• /
• v.8 no.2
• /
• pp.359-374
• /
• 2012

In this paper, we propose a new machine vision algorithm for automatic defect detection on patterned textures with the help of texture-periodicity and the Jensen-Shannon Divergence, which is a symmetrized and smoothed version of the Kullback-Leibler Divergence. Input defective images are split into several blocks of the same size as the size of the periodic unit of the image. Based on histograms of the periodic blocks, Jensen-Shannon Divergence measures are calculated for each periodic block with respect to itself and all other periodic blocks and a dissimilarity matrix is obtained. This dissimilarity matrix is utilized to get a matrix of true-metrics, which is later subjected to Ward's hierarchical clustering to automatically identify defective and defect-free blocks. Results from experiments on real fabric images belonging to 3 major wallpaper groups, namely, pmm, p2, and p4m with defects, show that the proposed method is robust in finding fabric defects with a very high success rates without any human intervention.

• Journal of Korea Society of Industrial Information Systems
• /
• v.8 no.4
• /
• pp.46-54
• /
• 2003

In this paper, we propose segmentation algorithm for MR brain images using the histogram of T1-weighted, T2-weighted and PD images. Segmentation algorithm is composed of 3 steps. The first step involves the extraction of cerebrum images by ram a cerebrum mask over three input images. In the second step, peak ranges are determined from the histogram of the cerebrum image. In the final step, cerebrum images are segmented using coarse to fine clustering technique. We compare the segmentation result and processing time according to peak ranges. Also compare with the other segmentation methods. The proposed algorithm achieved better segmentation results than the other methods.

• Journal of the Institute of Convergence Signal Processing
• /
• v.18 no.1
• /
• pp.21-24
• /
• 2017

Passive millimeter wave (MMW) imaging penetrates clothing to detect concealed objects. The distances extraction to the concealed objects is critical for the security and defense. In this paper, we address a passive stereo 3 mm MMW imaging system to extract the longitudinal distance to the concealed object. The concealed object area is segmented and extracted by the k-means clustering algorithm with splitting initialization. The distance to the concealed object is estimated by the corresponding centers of the segmented objects. In the experimental two pairs (each pair for horizontal and vertical polarization) of stereo MMW images are obtained to estimate distances to concealed objects.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2015.10a
• /
• pp.1597-1599
• /
• 2015

본 논문에서는 곡물이나 광석 등의 원료들 중에서 양품 및 불량품을 검출하기 위해, Color CCD 카메라로 촬영한 원료영상에서 Mean-Shift 클러스터링 알고리즘과 단계별 병합 방법을 제안하고 있다. 먼저 원료 학습 영상에서 배경을 제거하고 영상 색 분포정도를 기준으로 모폴로지를 이용하여 영상의 전경맵을 얻는다. 전경맵 영상에 대해서 Mean-Shift 군집화 알고리즘을 적용하여 영상을 N개의 군집으로 나누고, 단계별로 위치 근접성, 색상대푯값 유사성을 비교하여 비슷한 군집끼리 통합한다. 이렇게 통합된 원료 객체는 영상채널마다의 연관관계를 반영할 수 있도록 RG/GB/BR의 2차원 컬러분포도로 표현한다. 원료 객체별로 변환된 2차원 컬러 분포도에서 분포의 주성분의 기울기와 타원들을 생성한다. 객체별 분포 타원은 테스트 원료 영상데이터에서 양품과 불량품을 검출하는 임계값이 된다. 본 논문에서 제안한 방법으로 다양한 원료영상에 실험한 결과, 기존 선별방식에 비해 사용자의 인위적 조작이 적고 정확한 원료 선별 결과를 얻을 수 있었다.