• Journal of Broadcast Engineering
• /
• v.13 no.4
• /
• pp.544-553
• /
• 2008

With the development of video editing technology, there are growing uses of overlay text inserted into video contents to provide viewers with better visual understanding. Since the content of the scene or the editor's intention can be well represented by using inserted text, it is useful for video information retrieval and indexing. Most of the previous approaches are based on low-level features, such as edge, color, and texture information. However, existing methods experience difficulties in handling texts with various contrasts or inserted in a complex background. In this paper, we propose a novel framework to localize the overlay text in a video scene. Based on our observation that there exist transient colors between inserted text and its adjacent background a transition map is generated. Then candidate regions are extracted by using the transition map and overlay text is finally determined based on the density of state in each candidate. The proposed method is robust to color, size, position, style, and contrast of overlay text. It is also language free. Text region update between frames is also exploited to reduce the processing time. Experiments are performed on diverse videos to confirm the efficiency of the proposed method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.10
• /
• pp.1840-1848
• /
• 2017

Linear wireless sensor networks typically construct a network topology with a high reliability through sequential 1:1 mapping among sensor nodes, so that they are used in various surveillance applications of major national infrastructures. Most existing techniques for identifying sensor nodes in those networks are using GPS, AOA, and RSSI mechanisms. However, GPS or AOA based node identification techniques affect the size or production cost of the nodes so that it is not easy to construct practical sensor networks. RSSI based techniques may have a high deviation regrading location identification according to propagation environments and equipment quality so that complexity of error correction algorithm may increase. We propose a timestamps based sequential localization algorithm that uses transmit and receive timestamps in a message between sensor nodes without using GPS, AOA, and RSSI techniques. The algorithms for distance measurement between each node are expected to measure distance within up to 1 meter in case of an crystal oscillator of 300MHz or more.

• The KIPS Transactions:PartA
• /
• v.12A no.5 s.95
• /
• pp.451-460
• /
• 2005

We present a methodology to design energy-efficient data paths using FPGAs. Our methodology integrates domain specific modeling, coarse-grained performance evaluation, design space exploration, and low-level simulation to understand the tradeoffs between energy, latency, and area. The domain specific modeling technique defines a high-level model by identifying various components and parameters specific to a domain that affect the system-wide energy dissipation. A domain is a family of architectures and corresponding algorithms for a given application kernel. The high-level model also consists of functions for estimating energy, latency, and area that facilitate tradeoff analysis. Design space exploration(DSE) analyzes the design space defined by the domain and selects a set of designs. Low-level simulations are used for accurate performance estimation for the designs selected by the DSE and also for final design selection We illustrate our methodology using a family of architectures and algorithms for matrix multiplication. The designs identified by our methodology demonstrate tradeoffs among energy, latency, and area. We compare our designs with a vendor specified matrix multiplication kernel to demonstrate the effectiveness of our methodology. To illustrate the effectiveness of our methodology, we used average power density(E/AT), energy/(area x latency), as themetric for comparison. For various problem sizes, designs obtained using our methodology are on average $25\%$ superior with respect to the E/AT performance metric, compared with the state-of-the-art designs by Xilinx. We also discuss the implementation of our methodology using the MILAN framework.

• The KIPS Transactions:PartA
• /
• v.15A no.1
• /
• pp.1-8
• /
• 2008

An occlusion culling method, one of visibility culling methods, excludes invisible objects or triangles which are covered by other objects. As it reduces computation quantity, occlusion culling is an effective method to handle complex scenes in real-time. But an existing common occlusion culling method, such as hardware occlusion query method, sends objects' data twice to GPU and this causes processing overheads once for occlusion culling test and the other is for rendering. And another existing hardware occlusion culling method, VCBP, can test objects' visibility quickly, but it neither test bounding volume nor return test result to application stage. In this paper, we propose a single pass occlusion culling method which uses temporal and spatial coherency, with effective occlusion culling hardware architecture. In our approach, the hardware performs occlusion culling test rapidly with cache on the rasterization stage where triangles are transformed into fragments. At the same time, hardware sends each primitive's visibility information to application stage. As a result, the application stage reduces data transmission quantity by excluding covered objects using the visibility information on previous frame and hierarchical spatial tree. Our proposed method improved maximum 44%, minimum 14% compared with S&W method based on hardware occlusion query. And the performance is increased 25% and 17% respectively, compared to maximum and minimum performance of CHC method which is based on occlusion culling method.

• Journal of Broadcast Engineering
• /
• v.18 no.6
• /
• pp.835-849
• /
• 2013

Joint Collaborative Team on Video Coding (JCT-VC), which have founded ISO/IEC MPEG and ITU-T VCEG, has standardized High Efficiency Video Coding (HEVC). Standardization of HEVC has started with purpose of twice or more coding performance compared to H.264/AVC. However, flexible and hierarchical coding block and recursive coding structure are problems to overcome of HEVC standard. Many fast encoding algorithms for reducing computational complexity of HEVC encoder have been proposed. However, it is hard to implement a real-time HEVC encoder only with those fast encoding algorithms. In this paper, for implementation of software-based real-time HEVC encoder, data-level parallelism using SIMD instructions and CPU/GPU multi-threading methods are proposed. And we also proposed appropriate operations and functional modules to apply the proposed methods on HM 10.0 software. Evaluation of the proposed methods implemented on HM 10.0 software showed 20-30fps for $832{\times}480$ sequences and 5-10fps for $1920{\times}1080$ sequences, respectively.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.43 no.6 s.348
• /
• pp.9-17
• /
• 2006

We proposed a new High-speed CMOS Level Up/Down Shifter circuits that can be used with Dynamic Voltage and Frequency Scaling(DVFS) algorithm, for low power system in the SoC(System-on-Chip). This circuit used to interface between the other voltage levels in each CMOS circuit boundary, or between multiple core voltage levels in a system bus. Proposed circuit have advantage that decrease speed attenuation and duty ratio distortion problems for interface. The level up/down shifter of the proposed circuit designed that operated from multi core voltages$(0.6\sim1.6V)$ to used voltage level for each IP at the 500MHz input frequency The proposed circuit supports level up shifting from the input voltage levels, that are standard I/O voltages 1.8V, 2.5V, 3.3V, to multiple core voltage levels in between of $0.6V\sim1.6V$, that are used internally in the system. And level down shifter reverse operated at 1Ghz input frequency for same condition. Simulations results are shown to verify the proposed function by Hspice simulation, with $0.6V\sim1.6V$ CMOS Process, $0.13{\mu}m$ IBM CMOS Process and $0.65{\mu}m$ CMOS model parameters. Moreover, it is researched delay time, power dissipation and duty ration distortion of the output voltage witch is proportional to the operating frequency for the proposed circuit.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.22 no.5
• /
• pp.7-14
• /
• 2021

Tomato crops are highly affected by tomato diseases, and if not prevented, a disease can cause severe losses for the agricultural economy. Therefore, there is a need for a system that quickly and accurately diagnoses various tomato diseases. In this paper, we propose a system that classifies nine diseases as well as healthy tomato plants by applying various pretrained deep learning-based CNN models trained on an ImageNet dataset. The tomato leaf image dataset obtained from PlantVillage is provided as input to ResNet, Xception, and DenseNet, which have deep learning-based CNN architectures. The proposed models were constructed by adding a top-level classifier to the basic CNN model, and they were trained by applying a 5-fold cross-validation strategy. All three of the proposed models were trained in two stages: transfer learning (which freezes the layers of the basic CNN model and then trains only the top-level classifiers), and fine-tuned learning (which sets the learning rate to a very small number and trains after unfreezing basic CNN layers). SGD, RMSprop, and Adam were applied as optimization algorithms. The experimental results show that the DenseNet CNN model to which the RMSprop algorithm was applied output the best results, with 98.63% accuracy.

• Journal of Broadcast Engineering
• /
• v.24 no.3
• /
• pp.472-484
• /
• 2019

This paper proposes a time-domain sound event detection algorithm using DNN (Deep Neural Network). In this system, time domain sound waveform data which is not converted into the frequency domain is used as input to the DNN. The overall structure uses CRNN structure, and GLU, ResNet, and Squeeze-and-excitation blocks are applied. And proposed structure uses structure that considers features extracted from several layers together. In addition, under the assumption that it is practically difficult to obtain training data with strong labels, this study conducted training using a small number of weakly labeled training data and a large number of unlabeled training data. To efficiently use a small number of training data, the training data applied data augmentation methods such as time stretching, pitch change, DRC (dynamic range compression), and block mixing. Unlabeled data was supplemented with insufficient training data by attaching a pseudo-label. In the case of using the neural network and the data augmentation method proposed in this paper, the sound event detection performance is improved by about 6 %(based on the f-score), compared with the case where the neural network of the CRNN structure is used by training in the conventional method.

• Information Systems Review
• /
• v.24 no.1
• /
• pp.21-37
• /
• 2022

Drug repositioning, one of the methods of developing new drugs, is a useful way to discover new indications by allowing drugs that have already been approved for use in people to be used for other purposes. Recently, with the development of machine learning technology, the case of analyzing vast amounts of biological information and using it to develop new drugs is increasing. The use of machine learning technology to drug repositioning will help quickly find effective treatments. Currently, the world is having a difficult time due to a new disease caused by coronavirus (COVID-19), a severe acute respiratory syndrome. Drug repositioning that repurposes drugsthat have already been clinically approved could be an alternative to therapeutics to treat COVID-19 patients. This study intends to examine research trends in the field of drug repositioning using machine learning techniques. In Pub Med, a total of 4,821 papers were collected with the keyword 'Drug Repositioning'using the web scraping technique. After data preprocessing, frequency analysis, LDA-based topic modeling, random forest classification analysis, and prediction performance evaluation were performed on 4,419 papers. Associated words were analyzed based on the Word2vec model, and after reducing the PCA dimension, K-Means clustered to generate labels, and then the structured organization of the literature was visualized using the t-SNE algorithm. Hierarchical clustering was applied to the LDA results and visualized as a heat map. This study identified the research topics related to drug repositioning, and presented a method to derive and visualize meaningful topics from a large amount of literature using a machine learning algorithm. It is expected that it will help to be used as basic data for establishing research or development strategies in the field of drug repositioning in the future.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.8 no.1
• /
• pp.123-136
• /
• 2004

In this paper, complexity and regularity of polynomial multiplication over $GF({2^n})$ are defined by using Hamming weight of rows and columns of the matrix ever GF(2) which represents polynomial multiplication. It is shown experimentally that in order to construct the block cipher robust against differential cryptanalysis, polynomial multiplication of substitution layer and the permutation layer should have high complexity and high regularity. With result of the experiment, a way of constituting S box and G function is suggested in the block cipher whose structure is similar to SEED, which is KOREA standard of 128-bit block cipher. S box can be formed with a nonlinear function and an affine transform. Nonlinear function must be strong with differential attack and linear attack, and it consists of an inverse number over $GF({2^8})$ which has neither a fixed pout, whose input and output are the same except 0 and 1, nor an opposite fixed number, whose output is one`s complement of the input. Affine transform can be constituted so that the input/output correlation can be the lowest and there can be no fixed point or opposite fixed point. G function undergoes linear transform with 4 S-box outputs using the matrix of 4${\times}$4 over $GF({2^8})$. The components in the matrix of linear transformation have high complexity and high regularity. Furthermore, G function can be constituted so that MDS(Maximum Distance Separable) code can be formed, SAC(Strict Avalanche Criterion) can be met, and there can be no weak input where a fixed point an opposite fixed point, and output can be two`s complement of input. The primitive polynomials of nonlinear function affine transform and linear transformation are different each other. The S box and G function suggested in this paper can be used as a constituent of the block cipher with high security, in that they are strong with differential attack and linear attack with no weak input and they are excellent at diffusion.