• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• The KIPS Transactions:PartB
• /
• v.16B no.5
• /
• pp.341-346
• /
• 2009

The motion data glove is a representative human-computer interaction tool that inputs human hand gestures to computers by measuring their motions. The motion data glove is essential equipment used for new computer technologiesincluding home automation, virtual reality, biometrics, motion capture. For its popular usage, this paper attempts to develop an inexpensive visual.type motion data glove that can be used without any special equipment. The proposed approach has the special feature; it can be developed as a low-cost one becauseof not using high-cost motion-sensing fibers that were used in the conventional approaches. That makes its easy production and popular use possible. This approach adopts a visual method that is obtained by improving conventional optic motion capture technology, instead of mechanical method using motion-sensing fibers. Compared to conventional visual methods, the proposed method has the following advantages and originalities Firstly, conventional visual methods use many cameras and equipments to reconstruct 3D pose with eliminating occlusions But the proposed method adopts a mono vision approachthat makes simple and low cost equipments possible. Secondly, conventional mono vision methods have difficulty in reconstructing 3D pose of occluded parts in images because they have weak points about occlusions. But the proposed approach can reconstruct occluded parts in images by using originally designed thin-bar-shaped optic indicators. Thirdly, many cases of conventional methods use nonlinear numerical computation image analysis algorithm, so they have inconvenience about their initialization and computation times. But the proposed method improves these inconveniences by using a closed-form image analysis algorithm that is obtained from original formulation. Fourthly, many cases of conventional closed-form algorithms use approximations in their formulations processes, so they have disadvantages of low accuracy and confined applications due to singularities. But the proposed method improves these disadvantages by original formulation techniques where a closed-form algorithm is derived by using exponential-form twist coordinates, instead of using approximations or local parameterizations such as Euler angels.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.4
• /
• pp.144-159
• /
• 2014

Techniques for making a single panoramic image using multiple pictures are widely studied in many areas such as computer vision, computer graphics, etc. The panorama image can be applied to various fields like virtual reality, robot vision areas which require wide-angled shots as an useful way to overcome the limitations such as picture-angle, resolutions, and internal informations of an image taken from a single camera. It is so much meaningful in a point that a panoramic image usually provides better immersion feeling than a plain image. Although there are many ways to build a panoramic image, most of them are using the way of extracting feature points and matching points of each images for making a single panoramic image. In addition, those methods use the RANSAC(RANdom SAmple Consensus) algorithm with matching points and the Homography matrix to transform the image. The SURF(Speeded Up Robust Features) algorithm which is used in this paper to extract featuring points uses an image's black and white informations and local spatial informations. The SURF is widely being used since it is very much robust at detecting image's size, view-point changes, and additionally, faster than the SIFT(Scale Invariant Features Transform) algorithm. The SURF has a shortcoming of making an error which results in decreasing the RANSAC algorithm's performance speed when extracting image's feature points. As a result, this may increase the CPU usage occupation rate. The error of detecting matching points may role as a critical reason for disqualifying panoramic image's accuracy and lucidity. In this paper, in order to minimize errors of extracting matching points, we used $3{\times}3$ region's RGB pixel values around the matching points' coordinates to perform intermediate filtering process for removing wrong matching points. We have also presented analysis and evaluation results relating to enhanced working speed for producing a panorama image, CPU usage rate, extracted matching points' decreasing rate and accuracy.

• Journal of the Korea Computer Graphics Society
• /
• v.16 no.1
• /
• pp.1-7
• /
• 2010

With the widespread use of vision-based surveillance systems, the capability for person identification is now an essential component. However, the CCTV cameras used in surveillance systems tend to produce relatively low-resolution images, making it difficult to use face recognition techniques for person identification. Therefore, an algorithm is proposed for person identification in CCTV camera images based on the clothing. Whenever a person is authenticated at the main entrance of a building, the clothing feature of that person is extracted and added to the database. Using a given image, the clothing area is detected using background subtraction and skin color detection techniques. The clothing feature vector is then composed of textural and color features of the clothing region, where the textural feature is extracted based on a local edge histogram, while the color feature is extracted using octree-based quantization of a color map. When given a query image, the person can then be identified by finding the most similar clothing feature from the database, where the Euclidean distance is used as the similarity measure. Experimental results show an 80% success rate for person identification with the proposed algorithm, and only a 43% success rate when using face recognition.

• Korean Journal of Community Nutrition
• /
• v.20 no.3
• /
• pp.220-235
• /
• 2015

Objectives: The purpose of this study was to investigate how clinical nutrition services is provided at a long term care hospital in Korea and to investigate job satisfaction levels of the clinical dietitians. Methods: Survey questionnaire was sent to dietitians working at a long term care hospital in Korea. The participating hospitals (n=240) were randomly selected from 1,180 long- term care hospitals using a stratified sampling method. A total of 134 long term care hospital s and 223 dietitians completed the survey of clinical nutrition service s and job satisfaction questionnaires The job satisfaction questionnaire included 27 job satisfaction questions on task, stability vision, working conditions, and relationship areas. Results: The average nutritional screening rate was 17.9% and the rate of computerized nutritional screening system was 9.7% in the participating hospitals. Nutritional intervention rate was only 3.2% of all patients. KOIHA (Korea Institute for Healthcare Accreditation) accreditated hospitals showed only 50% performance rate of nutrition service evaluation area. This shows that after achieving KOIHA accredition, many hospitals do not emphasize the performance of nutritional services. The job satisfaction scores in all four areas ranged from 2/5 to 3/5, implying generally low job satisfaction level in hospital dietitians. Linear regression analysis results showed that the "hospital adequacy grade" type was a significant predictor of job satisfaction level for two areas (working conditions & relationship). Conclusions: There is a need to provide proper standardized clinical nutrition services as a primary treatment and we observed large variations in the quality of nutritional service s in long term care hospitals. Therefore, government and local hospitals have to work on implementing nutritional programs and policies for improved service and care.

• Journal of The Korean Society of Inherited Metabolic disease
• /
• v.16 no.2
• /
• pp.102-108
• /
• 2016

Mitochondrial encephalopathy, lactic acidosis, and stroke-like episode (MELAS) syndrome is one of mitochondrial encephalopathy. As the early clinical manifestations can be variable, it is important to suspect the disease, especially in patients with multiple organ dysfunctions. A boy was diagnosed with epilepsy when he was 9 years old. Two years later, severe headache and blurred vision developed suddenly. On examination, left homonymous hemianopsia was detected with corresponding cerebral parenchymal lesions in right temporo-occipito-parietal areas. MELAS syndrome was confirmed by genetic test, which showed m.3243 A>G mitochondrial DNA mutation. Multivitamins including coenzyme Q10 were added to anticonvulsant. He experienced 4 more events of stroke-like episodes over 5 years, but he is able to perform normal daily activities. A 13-year-old boy was brought to the hospital due to suddenly developed respiratory arrest and asystole associated with pneumonia. Past medical history revealed that he had multiple medical problems such as epilepsy, failure-to-thrive, optic atrophy, and deafness. He has been on valproic acid as an anticonvulsant which was prescribed from local clinic. He recovered after the resuscitation, but his cognition and motor function were severely damaged. He became bed-ridden. He was diagnosed with MELAS syndrome by brain MRI, muscle biopsy, and clinical features. Genetic test did not reveal any mitochondrial gene mutation. Four years later, he expired due to suddenly developed severe metabolic acidosis combined with hyperglycemic hyperosmolar nonketotic coma. The clinical features of MELAS syndrome are variable. Early diagnosis before the presentation to the grave clinical course may be important for the better clinical outcome.

• The Korean Journal of Applied Statistics
• /
• v.24 no.6
• /
• pp.1181-1196
• /
• 2011

Image fusion is the process of combining information from two or more source images of a scene into a single composite image with application to many fields, such as remote sensing, computer vision, robotics, medical imaging and defense. The most common wavelet-based fusion is discrete wavelet transform fusion in which the high frequency sub-bands and low frequency sub-bands are combined on activity measures of local windows such standard deviation and mean, respectively. However, discrete wavelet transform is not translation-invariant and it often yields block artifacts in a fused image. In this paper, we propose a robust image fusion based on the stationary wavelet transform to overcome the drawback of discrete wavelet transform. We use the activity measure of interquartile range as the robust estimator of variance in high frequency sub-bands and combine the low frequency sub-band based on the interquartile range information present in the high frequency sub-bands. We evaluate our proposed method quantitatively and qualitatively for image fusion, and compare it to some existing fusion methods. Experimental results indicate that the proposed method is more effective and can provide satisfactory fusion results.

• Korean Journal of Poultry Science
• /
• v.22 no.3
• /
• pp.167-178
• /
• 1995

The purpose of this research was to get basic information for the development of Korean native chicken industry by reviewing the current native chicken farm management and marketing situation of native chicken products(meat and eggs). The research was carried on the basis of the farm field survey covering 210 native chicken feeders out of 9 different local areas, and the results were as follows ; 1. Average raising size of native chicken flocks of sample farms was 1,787 heads and about 50% of those farms raised less than 500 heads chickens for self-sufficiency or on the side. 2. Most farmers made the decision to start on feeding native chickens in small scale with small amount of capital without sound feeding program, and their decision was mainly influenced by recommendation of mass-media( 19.5%) and neighbors (17.2%). 3. The average income per farm earned by raising the native chickens was 13,719 Won, and income per head of chicken was 8,800 Won. 4. About 40% of feeders expressed that the poor marketing management and lack of capital were the bottleneck to native chicken farm management. 5. About 70% of feeders evaluated the prospect of native chicken industry positively and so, about 60% of feeders hoped to expand the raising size in the future. 6. Most farmers directry made a bargain with marketer including middleman and enduser in selling the chicken products because there was not established special marketing system for native chicken products. 7. The sales age of native broiler was about 16~20 weeks and average body weight of broiler was 1.5~2.0 kg. And farm recieved price was not decided on the basis of each body weight or meat quality but only number of heads. 8. The average first egg-laying age of chickens was about 165 days and average annual laying rate was only about 56%. 9. In order to develop the successful Korean native chicken industry, followings are recommended ; 1) Reducing the production costs and increasing the productivity of native chickens should be carried out through technological research and development for sound feeding program of native chickens and sufficient fund supply. 2) Orderly native chicken marketing and pricing system should be established to give good vision about native chickens to farmers and to delight the consumers. 3) The measures for product differentiation including meat quality and nutritional value of native chicken products against other improved chickens should be actively taken by feeders and government.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.7
• /
• pp.496-503
• /
• 2020

The regeneration of Incheon Inner Port is a recent, representative case of conflict related to urban policies. This study aimed to analyze the conflicts that have arisen during the urban regeneration process. This study was a qualitative study, and the conflict management strategy was derived by analyzing the conflict process, subject, and content. The results of the analysis are as follows. First, central governmental agencies have proposed a clear plan that is mainly focused on the port redevelopment project through the participation of private sector businesses. Second, Incheon is pursuing a new vision called "Creative City" with specific urban regeneration. Third, the Incheon Port Authority is required to maximize the efficiency of the regeneration projects. Fourth, organizations such as port logistics companies and port trade unions are demanding the use of port space. Fifth, local residents and civic groups insist that the entire Inner Harbor should be returned to the citizens. Therefore, the establishment of city planning and administrative guidelines is necessary to manage Incheon Inner Harbor and surrounding areas in a desirable manner in order to develop a regeneration philosophy for Incheon Inner Harbor. Furthermore, the establishment of cooperative governance is required for the participation of various stakeholders.

• Journal of KIISE:Software and Applications
• /
• v.37 no.2
• /
• pp.104-109
• /
• 2010

The lighting reflection is a common problem in image analysis and causes the many difficulties to extract distinct features in related fields. Furthermore, the problem grows in the rainy night. In this paper, we aim to remove light reflection effects and reconstruct a road surface without lighting reflections in order to extract distinct features. The proposed method utilizes a 3D analysis based on a multiple geometry using captured images, with which we can combine each reflected areas; that is, we can remove lighting reflection effects and reconstruct the surface. At first, the regions of lighting sources and reflected surfaces are extracted by local maxima based on vertically projected intensity-histograms. After that, a fundamental matrix and homography matrix among multiple images are calculated by corresponding points in each image. Finally, we combine each surface by selecting minimum value among multiple images and replace it on a target image. The proposed method can reduces lighting reflection effects and the property on the surface is not lost. While the experimental results with collected data shows plausible performance comparing to the speed, reflection-overlapping areas which can not be reconstructed remain in the result. In order to solve this problem, a new reflection model needs to be constructed.