• Korean Journal of Remote Sensing
• /
• v.36 no.5_1
• /
• pp.807-821
• /
• 2020

The objective of this study is to evaluate the applicability of representative spatio-temporal fusion models developed for the fusion of mid- and low-resolution satellite images in order to construct a set of time-series high-resolution images for crop monitoring. Particularly, the effects of the characteristics of input image pairs on the prediction performance are investigated by considering the principle of spatio-temporal fusion. An experiment on the fusion of multi-temporal Sentinel-2 and RapidEye images in agricultural fields was conducted to evaluate the prediction performance. Three representative fusion models, including Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), SParse-representation-based SpatioTemporal reflectance Fusion Model (SPSTFM), and Flexible Spatiotemporal DAta Fusion (FSDAF), were applied to this comparative experiment. The three spatio-temporal fusion models exhibited different prediction performance in terms of prediction errors and spatial similarity. However, regardless of the model types, the correlation between coarse resolution images acquired on the pair dates and the prediction date was more significant than the difference between the pair dates and the prediction date to improve the prediction performance. In addition, using vegetation index as input for spatio-temporal fusion showed better prediction performance by alleviating error propagation problems, compared with using fused reflectance values in the calculation of vegetation index. These experimental results can be used as basic information for both the selection of optimal image pairs and input types, and the development of an advanced model in spatio-temporal fusion for crop monitoring.

• Sleep Medicine and Psychophysiology
• /
• v.14 no.1
• /
• pp.42-48
• /
• 2007

Introduction: Detrended fluctuation analysis (DFA) is used as a way of studying nonlinearity of EEG. In this study, DFA is applied on sleep EEG of normal subjects to look into its nonlinearity in terms of EEG channels and sleep stages. Method: Twelve healthy young subjects (age：$23.8{\pm}2.5$ years old, male：female=7：5) have undergone nocturnal polysomnography (nPSG). EEG from nPSG was classified in terms of its channels and sleep stages and was analyzed by DFA. Scaling exponents (SEs) yielded by DFA were compared using linear mixed model analysis. Results: Scaling exponents (SEs) of sleep EEG were distributed around 1 showing long term temporal correlation and self-similarity. SE of C3 channel was bigger than that of O1 channel. As sleep stage progressed from stage 1 to slow wave sleep, SE increased accordingly. SE of stage REM sleep did not show significant difference when compared with that of stage 1 sleep. Conclusion: SEs of Normal sleep EEG showed nonlinear characteristic with scale-free fluctuation, long-range temporal correlation, self-similarity and self-organized criticality. SE from DFA differentiated sleep stages and EEG channels. It can be a useful tool in the research with sleep EEG.

• Journal of KIISE:Databases
• /
• v.30 no.3
• /
• pp.237-251
• /
• 2003

In this paper, we propose a video retrieval system to search logical structure and semantic contents of video data efficiently. The proposed system employs a layered modelling method that orBanifes video data in raw data layer, content layer and key frame layer. The layered modelling of the proposed system represents logical structures and semantic contents of video data in content layer. Also, the proposed system supports various types of searches such as text search, visual feature based similarity search, spatio-temporal relationship based similarity search and semantic contents search.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• fall
• /
• pp.91-94
• /
• 2021

As deep learning technologies becoming developed, realistic fake videos synthesized by deep learning models called "Deepfake" videos became even more difficult to distinguish from original videos. As fake news or Deepfake blackmailing are causing confusion and serious problems, this paper suggests a novel model detecting Deepfake videos. We chose Residual Convolutional Neural Network (Resnet50) as an extraction model and Long Short-Term Memory (LSTM) which is a form of Recurrent Neural Network (RNN) as a classification model. We adopted cosine similarity with hinge loss to train our extraction model in embedding the features of Deepfake and original video. The result in this paper demonstrates that temporal features in the videos are essential for detecting Deepfake videos.

• Journal of Biomedical Engineering Research
• /
• v.29 no.4
• /
• pp.272-277
• /
• 2008

Functional MRI technique was used in this study for examining the language switching mechanisms between the first language (L1) and the second language (L2). Language switching mechanism is regarded as a complex task that involves an interaction between L1 and L2. The aim of study is to find out the brain activation patterns during the phonological process of reading real English words and English words written in Korean characters in a bilingual person. Korean-English bilingual subjects were examined while they covertly read four types of words native Korean words, Korean words of a foreign origin, English words written in Korean characters, and English words. The fMRI results reveal that the left hemispheric language-related regions at the brain, such as the left inferior frontal, superior temporal, and parietal cortices, have a greater response to the presentation of English words written in Korean characters than for the other types of words, in addition, a slight difference was observed in the occipital-temporal lobe. These results suggest that a change in the brain circuitry underlying the relational processes of language switching is mainly associated with general executive processing system in the left prefrontal cortex rather than with a similarity-based processing system in the occipital-temporal lobes.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.6 no.1
• /
• pp.426-445
• /
• 2012

Conventional rate control (RC) schemes for H.264 video coding usually regulate output bit rate to match channel bandwidth by adjusting quantization parameter (QP) at fixed full frame rate, and the passive frame skipping to avoid buffer overflow usually occurs when scene changes or high motions exist in video sequences especially at low bit rate, which degrades spatial-temporal quality and causes jerky effect. In this paper, an active content adaptive frame skipping scheme is proposed instead of passive methods, which skips subjectively trivial frames by structural similarity (SSIM) measurement between the original frame and the interpolated frame via motion vector (MV) copy scheme. The saved bits from skipped frames are allocated to coded key ones to enhance their spatial quality, and the skipped frames are well recovered based on MV copy scheme from adjacent key ones at the decoder side to maintain constant frame rate. Experimental results show that the proposed active SSIM-based frameskip scheme acquires better and more consistent spatial-temporal quality both in objective (PSNR) and subjective (SSIM) sense with low complexity compared to classic fixed frame rate control method JVT-G012 and prior objective metric based frameskip method.

• Journal of KIISE:Software and Applications
• /
• v.30 no.7_8
• /
• pp.610-617
• /
• 2003

Similarity matching in video database is of growing importance in many new applications such as video clustering and digital video libraries. In order to provide efficient access to relevant data in large databases, there have been many research efforts in video indexing with diverse spatial and temporal features. however, most of the previous works relied on sequential matching methods or memory-based inverted file techniques, thus making them unsuitable for a large volume of video databases. In order to resolve this problem, this paper proposes an effective and scalable indexing technique using a trie, originally proposed for string matching, as an index structure. For building an index, we convert each frame into a symbol sequence using a window order heuristic and build a disk-resident trie from a set of symbol sequences. For query processing, we perform a depth-first search on the trie and execute a temporal segmentation. To verify the superiority of our approach, we perform several experiments with real and synthetic data sets. The results reveal that our approach consistently outperforms the sequential scan method, and the performance gain is maintained even with a large volume of video databases.

• Korean Journal of Remote Sensing
• /
• v.38 no.4
• /
• pp.327-341
• /
• 2022

Cloud removal is often required to construct time-series sets of optical images for environmental monitoring. In regression-based cloud removal, the selection of an appropriate regression model and the impact analysis of the input images significantly affect the prediction performance. This study evaluates the potential of Gaussian process (GP) regression for cloud removal and also analyzes the effects of cloud-free optical images and spectral bands on prediction performance. Unlike other machine learning-based regression models, GP regression provides uncertainty information and automatically optimizes hyperparameters. An experiment using Sentinel-2 multi-spectral images was conducted for cloud removal in the two agricultural regions. The prediction performance of GP regression was compared with that of random forest (RF) regression. Various combinations of input images and multi-spectral bands were considered for quantitative evaluations. The experimental results showed that using multi-temporal images with multi-spectral bands as inputs achieved the best prediction accuracy. Highly correlated adjacent multi-spectral bands and temporally correlated multi-temporal images resulted in an improved prediction accuracy. The prediction performance of GP regression was significantly improved in predicting the near-infrared band compared to that of RF regression. Estimating the distribution function of input data in GP regression could reflect the variations in the considered spectral band with a broader range. In particular, GP regression was superior to RF regression for reproducing structural patterns at both sites in terms of structural similarity. In addition, uncertainty information provided by GP regression showed a reasonable similarity to prediction errors for some sub-areas, indicating that uncertainty estimates may be used to measure the prediction result quality. These findings suggest that GP regression could be beneficial for cloud removal and optical image reconstruction. In addition, the impact analysis results of the input images provide guidelines for selecting optimal images for regression-based cloud removal.

• Journal of the Korean Society of Radiology
• /
• v.17 no.7
• /
• pp.1041-1047
• /
• 2023

Temporal bone CT, which is a high-resolution CT, uses a high tube voltage and a thin section thickness, so the scan dose is higher than that of adjacent areas. Accordingly, we applied changes to the reconstruction algorithm among the test conditions to find an algorithm with excellent sensitivity to lesions while reducing the test dose, and investigated its significance and the possibility of providing basic clinical data. As a result, when the tube voltage was lowered to 100 kVp and applied, the dose was reduced by about 35.6%, and when the definition algorithm was applied to the raw data acquired at 100 kVp, the SNR and CNR were excellent, and a statistically significant difference was shown when compared to other algorithms(p<0.05). And as a result of comparing structural similarity, the SSIM index was analyzed as 0.776, 0.813, and 0.741 for each ROI. Therefore, we believe that applying algorithm changes to temporal bone CT scans can partially reduce the dose generated from CT scans and are very meaningful in terms of basic clinical data.

• Atmosphere
• /
• v.25 no.2
• /
• pp.193-203
• /
• 2015

A large aperture scintillometer (LAS) was deployed with an optical path length of 2.1 km to estimate turbulent sensible heat flux (${\mathcal{Q}}_H$) over a highly heterogeneous urban area. Scintillation measurements were conducted during cold season in November and December 2013, and the daytime data of 14 days were used in the analysis after quality control processes. The LAS-derived ${\mathcal{Q}}_H$ show reasonable temporal variation ranging $20{\sim}160W\;m^{-2}$ in unstable atmospheric conditions, and well compare with the measured net radiation. The LAS footprint analysis suggests that ${\mathcal{Q}}_H$ can be relatively high when the newly built-up urban area has high source contribution of the turbulent flux in the study area ('northwesterly winds'). Sensitivity tests show that the LAS-derived ${\mathcal{Q}}_H$ are highly sensitive to non-dimensional similarity function for temperature structure function parameter, but relatively less sensitive to surface aerodynamic parameters and meteorological variables (temperature and wind speed). A lower Bowen ratio also has a significant influence on the flux estimation. Overall uncertainty of the estimated daytime ${\mathcal{Q}}_H$ is expected within about 20% at an upper limit for the analysis data. It is also found that stable atmospheric conditions can be poorly determined when the scintillometry technique is applied over the highly heterogeneous urban area.