• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.3
• /
• pp.877-893
• /
• 2022

With the development of deep learning, face inpainting has been significantly enhanced in the past few years. Although image inpainting framework integrated with generative adversarial network or attention mechanism enhanced the semantic understanding among facial components, the issues of reconstruction on corrupted regions are still worthy to explore, such as blurred edge structure, excessive smoothness, unreasonable semantic understanding and visual artifacts, etc. To address these issues, we propose a Learnable Structure Knowledge of Fusion Network (LSK-FNet), which learns a prior knowledge by edge generation network for image inpainting. The architecture involves two steps: Firstly, structure information obtained by edge generation network is used as the prior knowledge for face inpainting network. Secondly, both the generated prior knowledge and the incomplete image are fed into the face inpainting network together to get the fusion information. To improve the accuracy of inpainting, both of gated convolution and region normalization are applied in our proposed model. We evaluate our LSK-FNet qualitatively and quantitatively on the CelebA-HQ dataset. The experimental results demonstrate that the edge structure and details of facial images can be improved by using LSK-FNet. Our model surpasses the compared models on L1, PSNR and SSIM metrics. When the masked region is less than 20%, L1 loss reduce by more than 4.3%.

• Mycobiology
• /
• v.49 no.6
• /
• pp.559-581
• /
• 2021

The coastal marine ecosystems of Vietnam are one of the global biodiversity hotspots, but the biodiversity of marine fungi is not well known. To fill this major gap of knowledge, we assessed the genetic diversity (ITS sequence) of 75 fungal strains isolated from 11 surface coastal marine and deeper waters in Nha Trang Bay and Van Phong Bay using a culture-dependent approach and 5 OTUs (Operational Taxonomic Units) of fungi in three representative sampling sites using next-generation sequencing. The results from both approaches shared similar fungal taxonomy to the most abundant phylum (Ascomycota), genera (Candida and Aspergillus) and species (Candida blankii) but were different at less common taxa. Culturable fungal strains in this study belong to 3 phyla, 5 subdivisions, 7 classes, 12 orders, 17 families, 22 genera and at least 40 species, of which 29 species have been identified and several species are likely novel. Among identified species, 12 and 28 are new records in global and Vietnamese marine areas, respectively. The analysis of enzyme activity and the checklist of trophic mode and guild assignment provided valuable additional biological information and suggested the ecological function of planktonic fungi in the marine food web. This is the largest dataset of marine fungal biodiversity on morphology, phylogeny and enzyme activity in the tropical coastal ecosystems of Vietnam and Southeast Asia. Biogeographic aspects, ecological factors and human impact may structure mycoplankton communities in such aquatic habitats.

• The Bulletin of The Korean Astronomical Society
• /
• v.46 no.2
• /
• pp.76.2-76.2
• /
• 2021

We present the characterization of the performance of the next-generation controller (SDSU Gen III) for BOAO Echelle Spectrograph CCD (BOES CCD) at the Bohyunsan Optical Astronomy Observatory. The current controller (SDSU Gen II) of the BOES CCD will be upgraded to SDSU Gen III to provide a more stabilized operation. To assess the performance of the new controller (e.g., conversion gain, full well capacity, S/N), we obtain various types of calibration images (e.g., bias, flat, science images of standard stars). Based on those datasets, we find that the overall performance of the new controller is somewhat comparable to that of the old controller if the slow mode is adopted for the readout. This may demonstrate that the new controller can be successfully substituted for the old controller without a substantial loss of performance. However, further analysis with a large dataset obtained in various observational conditions is necessary to confirm our results.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.10a
• /
• pp.72-74
• /
• 2021

Recently, the spread of deep learning environments has increased the importance of dataset generation. Therefore, data is being augmented using GAN for efficient data set generation. However, several problems have been found in data generated using GAN, such as problems that occur in the early stages of learning and pixel breakage occurring in the generated image. In this paper, we intend to implement an image data selection algorithm to solve various problems arising from the existing GAN. The broken image screening algorithm was implemented to analyze the histogram distribution in the image and determine whether to store the generated image according to whether the result value satisfies the specified threshold value.

• Electronics and Telecommunications Trends
• /
• v.33 no.3
• /
• pp.89-99
• /
• 2018

Human activity recognition (HAR) is a technology that aims to offer an automatic recognition of what a person is doing with respect to their body motion and gestures. HAR is essential in many applications such as human-computer interaction, health care, rehabilitation engineering, video surveillance, and artificial intelligence. Smartphones are becoming the most popular platform for activity recognition owing to their convenience, portability, and ease of use. The noticeable change in smartphone-based activity recognition is the adoption of a deep learning algorithm leading to successful learning outcomes. In this article, we analyze the technology trend of activity recognition using smartphone sensors, challenging issues for future development, and a strategy change in terms of the generation of a activity recognition dataset.

• Asia pacific journal of information systems
• /
• v.16 no.1
• /
• pp.127-144
• /
• 2006

The K-means algorithm is widely used at the initial stage of data analysis in data mining process, partly because of its low time complexity and the simplicity of practical implementation. Cluster validity indices are used along with the algorithm in order to determine the number of clusters as well as the clustering results of datasets. In this paper, we present a performance comparison of sixteen indices, which are selected from forty indices in literature, while considering their applicability to nonhierarchical clustering algorithms. Data sets used in the experiment are generated based on multivariate normal distribution. In particular, four error types including standardization, outlier generation, error perturbation, and noise dimension addition are considered in the comparison. Through the experiment the effects of varying number of points, attributes, and clusters on the performance are analyzed. The result of the simulation experiment shows that Calinski and Harabasz index performs the best through the all datasets and that Davis and Bouldin index becomes a strong competitor as the number of points increases in dataset.

• Proceedings of the KSRS Conference
• /
• 2007.10a
• /
• pp.122-125
• /
• 2007

In recent years, there has been an increasing demand for improved accuracy and reliability of Earth Observation Satellite (EOS) data. Most of the data users in the field of remote sensing require understanding of product accuracy and uncertainty. Especially, EOS application products should be validated for practical application in the field. In order to evaluate the availability and applicability of application products, it will be necessary to establish a systematic validation system including techniques, equipments, ground truth data, etc. The Product Validation Site (PVS) for generation and validation of KOMPSAT application products was designed and established with various in-situ equipment and dataset. This paper presents the status of PVS and summarizes some results from experiment studies at PVS.

• IEIE Transactions on Smart Processing and Computing
• /
• v.4 no.2
• /
• pp.110-114
• /
• 2015

This paper reports a novel object-detection technique in video sequences. The proposed algorithm consists of detection of objects in illuminated and occluded videos by using object features and a neural network technique. It consists of two functional modules: region-based object feature extraction and continuous detection of objects in video sequences with region features. This scheme is proposed as an enhancement of the Lowe's scale-invariant feature transform (SIFT) object detection method. This technique solved the high computation time problem of feature generation in the SIFT method. The improvement is achieved by region-based feature classification in the objects to be detected; optimal neural network-based feature reduction is presented in order to reduce the object region feature dataset with winner pixel estimation between the video frames of the video sequence. Simulation results show that the proposed scheme achieves better overall performance than other object detection techniques, and region-based feature detection is faster in comparison to other recent techniques.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.38 no.1
• /
• pp.74-82
• /
• 2015

Two key challenges in statistical relational learning are uncertainty and complexity. Standard frameworks for handling uncertainty are probability and first-order logic respectively. A Markov logic network (MLN) is a first-order knowledge base with weights attached to each formula and is suitable for classification of dataset which have variables correlated with each other. But we need domain knowledge to construct first-order logics and a computational complexity problem arises when calculating weights of first-order logics. To overcome these problems we suggest a method to generate first-order logics and learn weights using association analysis in this study.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.12 no.2
• /
• pp.108-112
• /
• 2012

A sustainable production of electricity is essential for low carbon green growth in South Korea. The generation of wind power as renewable energy has been rapidly growing around the world. Undoubtedly wind energy is unlimited in potential. However, due to its own intermittency and volatility, there are difficulties in the effective harvesting of wind energy and the integration of wind power into the current electric power grid. To cope with this, many works have been done for wind speed and power forecasting. It is reported that, compared with physical persistent models, statistical techniques and computational methods are more useful for short-term forecasting of wind power. Among them, support vector regression (SVR) has much attention in the literature. This paper proposes an SVR based wind speed forecasting. To improve the forecasting accuracy, a fuzzy clustering is adopted in the process of SVR modeling. An illustrative example is also given by using real-world wind farm dataset. According to the experimental results, it is shown that the proposed method provides better forecasts of wind power.