• The KIPS Transactions:PartB
• /
• v.9B no.4
• /
• pp.509-516
• /
• 2002

This paper proposes an efficient algorithm which recognizes the mixed document image consisting of the images, texts, tables, and straight lines. This system is composed of three steps. The first step is the detection of rotation angle for complementing skewed images, the second is detection of erasing an unnecessary background region and last is the classification of each component included in document images. This algorithm performs preprocessing of detecting rotation angles and correcting documents based on the detected rotation angles in order to minimize the error rate by skewness of the documentation. We detected the rotation angie using only horizontal and vertical components in document images and minimized calculation time by erasing unnecessary background region in the detecting process of component of document. In the next step, we classify various components such as image, text, table and line area included in document images. we applied this method to various document images in order to evaluate the performance of document recognition system and show the successful experimental results.

• Journal of Broadcast Engineering
• /
• v.18 no.4
• /
• pp.619-630
• /
• 2013

In this paper, we propose an edge directed color demosaicing algorithm considering color channel correlation. The proposed method consists of local region classification step and edge directional interpolation step. In the first step, each region of a given Bayer image is classified as normal edge, pattern edge, and flat regions by using intra channel and inter channel gradients. Especially, two criteria and verification process for the normal edge and pattern edge classification are used to reduce edge direction estimation error, respectively. In the second step, edge directional interpolation process is performed according to characteristics of the classified regions. For horizontal and vertical directional interpolations, missing color components are obtained from interpolation equations based on intra channel and inter channel correlations in order to improve the performance of the directional interpolations. The simulation results show that the proposed algorithm outperforms conventional approaches in both objective and subjective terms.

• Journal of Korean Society for Geospatial Information Science
• /
• v.19 no.3
• /
• pp.41-47
• /
• 2011

Processing of the raw LiDAR data requires the high-end processor, because data form is a vector. In contrast, if LiDAR data is converted into a regular grid pattern by filltering, that has advantage of being in a low-cost equipment, because of the simple structure and faster processing speed. Especially, by using grid data classification, such as Quadtree, some of trees and cars are removed, so it has advantage of modeling. Therefore, this study presents the algorithm for automatic extraction of ground points using Quadtree and refion growing method from LiDAR data. In addition, Error analysis was performed based on the 1:5000 digital map of sample area to analyze the classification of ground points. In a result, the ground classification accuracy is over 98%. So it has the advantage of extracting the ground points. In addition, non-ground points, such as cars and tree, are effectively removed as using Quadtree and region growing method.

• Survey Research
• /
• v.4 no.2
• /
• pp.1-21
• /
• 2003

We proposed imputation strategies for marital status in the Population and Housing Census 2000 in Korea to illustrate the effective missing value imputation methods for social survey. The marital status which have relatively high non-response rates in the Census are considered to develope the effective missing value imputation procedures. The Classification and Regression Tree(CART)is employed to construct the imputation cells for hot-deck imputation, as well as to predict the missing value by model-based approach. We compare to imputation methods which include the CART model-based imputation and the sequential hot-deck imputation based on CART. Also we check whether different modeling for each region provides the more improved results. The results suggest that the proposed hot-deck imputation based on CART is very efficient and strongly recommendable. And the results show that different modeling for each region is not necessary.

• Journal of Broadcast Engineering
• /
• v.27 no.4
• /
• pp.538-547
• /
• 2022

Recently introduced image classification methods using Transformers show remarkable performance improvements over conventional neural network-based methods. In order to effectively consider regional features, research has been actively conducted on how to apply transformers by dividing image areas into multiple window areas, but learning of inter-window relationships is still insufficient. In this paper, to overcome this problem, we propose a transformer structure that can reflect the relationship between windows in learning. The proposed method computes the importance of each window region through compression and a fully connected layer based on self-attention operations for each window region. The calculated importance is scaled to each window area as a learned weight of the relationship between the window areas to re-calibrate the feature value. Experimental results show that the proposed method can effectively improve the performance of existing transformer-based methods.

• Journal of Information Processing Systems
• /
• v.14 no.5
• /
• pp.1176-1194
• /
• 2018

The object classification using the images' contents is a big challenge in computer vision. The superpixels' information can be used to detect and classify objects in an image based on locations. In this paper, we proposed a methodology to detect and classify the image's pixels' locations using enhanced bag of words (BOW). It calculates the initial positions of each segment of an image using superpixels and then ranks it according to the region score. Further, this information is used to extract local and global features using a hybrid approach of Scale Invariant Feature Transform (SIFT) and GIST, respectively. To enhance the classification accuracy, the feature fusion technique is applied to combine local and global features vectors through weight parameter. The support vector machine classifier is a supervised algorithm is used for classification in order to analyze the proposed methodology. The Pascal Visual Object Classes Challenge 2007 (VOC2007) dataset is used in the experiment to test the results. The proposed approach gave the results in high-quality class for independent objects' locations with a mean average best overlap (MABO) of 0.833 at 1,500 locations resulting in a better detection rate. The results are compared with previous approaches and it is proved that it gave the better classification results for the non-rigid classes.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.43 no.5 s.311
• /
• pp.38-47
• /
• 2006

Conventional classification method uses a single classifier based on shape or motion feature. However this method exhibits a weakness if naively used since the classification performance is highly sensitive to the accuracy of moving region to be detected. The detection accuracy, in turn, depends on the condition of the image background. In this paper, we propose to resolve the drawback and thus strengthen the classification reliability by employing a Bayesian decision fusion and by optimally combining the decisions of three classifiers. The first classifier is based on shape information obtained from Fourier descriptors while the second is based on the shape information obtained from image gradients. The third classifier uses motion information. Our experimental results on the classification Performance of human and vehicle with a static camera in various directions confirm a significant improvement and indicate the superiority of the proposed decision fusion method compared to the conventional Majority Voting and Weight Average Score approaches.

• Asian-Australasian Journal of Animal Sciences
• /
• v.29 no.5
• /
• pp.624-630
• /
• 2016

Many studies have reported the frequency and distribution of haplogroups among various cattle breeds for verification of their origins and genetic diversity. In this study, 318 complete sequences of the mtDNA control region from four Korean cattle breeds were used for haplogroup classification. 71 polymorphic sites and 66 haplotypes were found in these sequences. Consistent with the genetic patterns in previous reports, four haplogroups (T1, T2, T3, and T4) were identified in Korean cattle breeds. In addition, T1a, T3a, and T3b sub-haplogroups were classified. In the phylogenetic tree, each haplogroup formed an independent cluster. The frequencies of T3, T4, T1 (containing T1a), and T2 were 66%, 16%, 10%, and 8%, respectively. Especially, the T1 haplogroup contained only one haplotype and a sample. All four haplogroups were found in Chikso, Jeju black and Hanwoo. However, only the T3 and T4 haplogroups appeared in Heugu, and most Chikso populations showed a partial of four haplogroups. These results will be useful for stable conservation and efficient management of Korean cattle breeds.

• The KIPS Transactions:PartB
• /
• v.13B no.4 s.107
• /
• pp.383-392
• /
• 2006

In this paper, we propose a new relevance feedback algorithm using Probabilistic Neural Networks(PNN) while supporting multi-class learning. Then, to validate the effectiveness of our feedback approach, we incorporate the proposed algorithm into our region-based image retrieval tool, FRIP(Finding Regions In the Pictures). In our feedback approach, there is no need to assume that feature vectors are independent, and as well as it allows the system to insert additional classes for detail classification. In addition, it does not have a long computation time for training because it only has four layers. In the PNN classification process, we store the user's entire past feedback actions as a history in order to improve performance for future iterations. By using a history, our approach can capture the user's subjective intension more precisely and prevent retrieval performance errors which originate from fluctuating or degrading in the next iteration. The efficacy of our method is validated using a set of 3000 images derived from a Corel-photo CD.

• The Bulletin of The Korean Astronomical Society
• /
• v.39 no.2
• /
• pp.50-50
• /
• 2014

We present a new catalog of galaxies in the wider region of the Virgo cluster, based on the Sloan Digital Sky Survey (SDSS) Data Release 7. The Extended Virgo Cluster Catalog (EVCC) covers an area of 725 deg2 or 60.1 Mpc2. It is 5.2 times larger than the footprint of the classical Virgo Cluster Catalog (VCC) and reaches out to 3.5 times the virial radius of the Virgo cluster. We selected 1324 spectroscopically targeted galaxies with radial velocities less than 3000 km s-1. In addition, 265 galaxies that have been missed in the SDSS spectroscopic survey but have available redshifts in the NASA Extragalactic Database are also included. Our selection process secured a total of 1589 galaxies of which 676 galaxies are not included in the VCC. The certain and possible cluster members are defined by means of redshift comparison with a cluster infall model. We employed two independent and complementary galaxy classification schemes: the traditional morphological classification based on the visual inspection of optical images and a characterization of galaxies from their spectroscopic features. SDSS u, g, r, i, and z passband photometry of all EVCC galaxies was performed using Source Extractor. We compare the EVCC galaxies with the VCC in terms of morphology, spatial distribution, and luminosity function. The EVCC defines a comprehensive galaxy sample covering a wider range in galaxy density that is significantly different from the inner region of the Virgo cluster. It will be the foundation for forthcoming galaxy evolution studies in the extended Virgo cluster region, complementing ongoing and planned Virgo cluster surveys at various wavelengths.