• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.200-207
• /
• 2020

Recently, deep neural networks such as CNN are showing excellent performance in various fields such as image classification, object recognition, visual quality enhancement, etc. However, as the model size and computational complexity of deep learning models for most applications increases, it is hard to apply neural networks to IoT and mobile environments. Therefore, neural network compression algorithms for reducing the model size while keeping the performance have been being studied. In this paper, we apply few compression methods to CNN models and evaluate their performances in the embedded environment. For evaluate the performance, the classification performance and inference time of the original CNN models and the compressed CNN models on the image inputted by the camera are evaluated in the embedded board equipped with QCS605, which is a customized AI chip. In this paper, a few CNN models of MobileNetV2, ResNet50, and VGG-16 are compressed by applying the methods of pruning and matrix decomposition. The experimental results show that the compressed models give not only the model size reduction of 1.3~11.2 times at a classification performance loss of less than 2% compared to the original model, but also the inference time reduction of 1.2~2.21 times, and the memory reduction of 1.2~3.8 times in the embedded board.

• Journal of KIISE:Computer Systems and Theory
• /
• v.32 no.1
• /
• pp.29-38
• /
• 2005

The fluid animation procedure consists of physical simulation and visual rendering. In the physical simulation of fluids, the most frequently used practices are the numerical simulation of fluid particles using particle dynamics equations and the continuum analysis of flow via Wavier-Stokes equation. Particle dynamics method is fast in calculation, but the resulting fluid motion is conditionally unrealistic The method using Wavier-Stokes equation, on the contrary, yields lifelike fluid motion when properly conditioned, yet the complexity of calculation restrains this method from being used in real-time applications. Global illumination is generally successful in producing premium-Duality rendered images, but is also excessively slow for real-time applications. In this paper, we propose a rapid fluid animation method incorporating enhanced particle dynamics simulation method and pre-integrated volume rendering technique. The particle dynamics simulation of fluid flow was conducted in real-time using Lennard-Jones model, and the computation efficiency was enhanced such that a small number of particles can represent a significant volume. For real-time rendering, pre-integrated volume rendering method was used so that fewer slices than ever can construct seamless inter-laminar shades. The proposed method could successfully simulate and render the fluid motion in real time at an acceptable speed and visual quality.

• Korean Journal of Remote Sensing
• /
• v.36 no.5_1
• /
• pp.653-666
• /
• 2020

Understanding the vegetation structure is important to manage forest resources for sustainable forest development. With the recent development of technology, it is possible to apply new technologies such as drones and deep learning to forests and use it to estimate the vegetation structure. In this study, the vegetation structure of Gongju, Samchuk, and Seoguipo area was identified by fusion of drone-optical images and LiDAR data using Artificial Neural Networks(ANNs) with the accuracy of 92.62% (Kappa value: 0.59), 91.57% (Kappa value: 0.53), and 86.00% (Kappa value: 0.63), respectively. The vegetation structure analysis technology using deep learning is expected to increase the performance of the model as the amount of information in the optical and LiDAR increases. In the future, if the model is developed with a high-complexity that can reflect various characteristics of vegetation and sufficient sampling, it would be a material that can be used as a reference data to Korea's policies and regulations by constructing a country-level vegetation structure map.

• Korean Journal of Remote Sensing
• /
• v.36 no.5_1
• /
• pp.775-784
• /
• 2020

Recent global warming and anthropogenic activities have caused more frequent and massive wildfires with longer durations and more significant damages. MODIS has been monitoring global wildfires for almost 20 years, and GK2A and Himawari-8 are observing the wildfires in East Asia 144 times a day. However, the spatial resolution of 1 to 2 km is not sufficient for the detection of small and medium-size active fires, and therefore the studies on the active fire detection using high-resolution images are essential. However, there is no official product for the high-resolution active fire detection. Hence, we implemented the active fire detection algorithm of Landsat 8 and carried out a high-resolution-based detection of active fires in Australia in 2019, followed by the comparisons with the products of Himawari-8 and MODIS. Regarding the intense fires, the three satellites showed similar results, whereas the weak igniting and extinguishing fires or the fires in narrow areas were detected by only Landsat 8 with a 30m resolution. Small-sized fires, which are the majority in Korea, can be detected by the high-resolution satellites such as Landsat 8, Sentinel-2, Kompsat-3A, and the forthcoming Kompsat-7. Also, a comprehensive analysis together with the geostationary satellites in East Asia such as GK2A, Himawari-8, and Fengyun-3 will help the interoperability and the improvement of spatial and temporal resolutions.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.12
• /
• pp.2213-2220
• /
• 2017

This paper proposes an efficient approach for hardware implementation of moving object detection (MOD) processor using effective Gaussian mixture learning (EGML)-based background subtraction method. Arithmetic units used in background generation were implemented using LUT-based approximation to reduce hardware complexity. Hardware resources used for both background subtraction and Gaussian probability density calculation were shared. The MOD processor was verified by FPGA-in-the-loop simulation using MATLAB/Simulink. The MOD performance was evaluated by using six types of video defined in IEEE CDW-2014 dataset, which resulted the average of recall value of 0.7700, the average of precision value of 0.7170, and the average of F-measure value of 0.7293. The MOD processor was implemented with 882 slices and block RAM of $146{\times}36kbits$ on Virtex5 FPGA, resulting in 60% hardware reduction compared to conventional design based on EGML. It was estimated that the MOD processor could operate with 75 MHz clock, resulting in real-time processing of $800{\times}600$ video with a frame rate of 39 fps.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.49 no.12
• /
• pp.156-163
• /
• 2012

Recently, Information Hiding Technology are becoming increasingly demanding in the field of international security, military and medical image This paper proposes data hiding technique utilizing parity checker for gray level image. many researches have been adopted LSB substitution and XOR operation in the field of steganography for the low complexity, high embedding capacity and high image quality. But, LSB substitution methods are not secure through it's naive mechanism even though it achieves high embedding capacity. Proposed method replaces LSB of each pixel with XOR(between the parity check bit of other 7 MSBs and 1 Secret bit) within one pixel. As a result, stego-image(that is, steganogram) doesn't result in high image degradation. Eavesdropper couldn't easily detect the message embedding. This approach is applying the concept of symmetric-key encryption protocol onto steganography. Furthermore, 1bit of symmetric-key is generated by the self-reference of each pixel. Proposed method provide more 25% embedding rate against existing XOR operation-based methods and show the effect of the reversal rate of LSB about 2% improvement.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.44 no.4
• /
• pp.55-62
• /
• 2007

This paper proposes a defective pixel detection algorithm and its hardware structure for CCD/CMOS image sensor. In previous algorithms, the characteristics of image have not been considered. Also, some algorithms need quite a time to detect defective pixels. In order to make up for those disadvantages, the proposed defective pixel detection method detects defective pixels efficiently by considering the edges in the image and verifies them using several frames while checking scene-changes. Whenever scene-change is occurred, potentially defective pixels are checked and confirmed whether it is defective or not. Test results showed that the correct detection rate in a frame was increased 6% and the defective pixel verification time was decreased 60%. The proposed algorithm was implemented with verilog HDL. The edge indicator in color interpolation block was reused. Total logic gate count was 5.4k using 0.25um CMOS standard cell library.

• Journal of the Institute of Convergence Signal Processing
• /
• v.8 no.4
• /
• pp.237-243
• /
• 2007

In this paper, we propose the Real-time Dead-Pixel Detection and Compensation System for mobile camera and its hardware architecture. The CMOS image sensors as image input devices are becoming popular due to the demand for miniaturized, low-power and cost-effective imaging systems. However a conventional Dead-Pixel Detection Algorithm is disable to detect neighboring dead pixels and it degrades image quality by wrong detection and compensation. To detect dead pixels the proposed system is classifying dead pixels into Hot pixel and Cold pixel. Also, the proposed algorithm is processing line-detector and $5{\times}5$ window-detector consecutively. The line-detector and window-detector can search dead pixels by using one-dimensional(only horizontal) method in low frequency area and two-dimensional(vertical and diagonal) method in high frequency area, respectively. The experimental result shows that it can detect 99% of dead pixels. It was designed in Verilog hardware description language and total gate count is 23K using TSMC 0.25um ASIC library.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.7C
• /
• pp.675-686
• /
• 2005

In this paper, we propose an efficient face region detection technique for the content-based video summarization. To segment video, shot changes are detected from a video sequence and key frames are selected from the shots. We select one frame that has the least difference between neighboring frames in each shot. The proposed face detection algorithm detects face region from selected key frames. And then, we provide user with summarized frames included face region that has an important meaning in dramas or movies. Using Bayes classification rule and statistical characteristic of the skin pixels, face regions are detected in the frames. After skin detection, we adopt the projection method to segment an image(frame) into face region and non-face region. The segmented regions are candidates of the face object and they include many false detected regions. So, we design a classifier to minimize false lesion using CART. From SGLD matrices, we extract the textual feature values such as Inertial, Inverse Difference, and Correlation. As a result of our experiment, proposed face detection algorithm shows a good performance for the key frames with a complex and variant background. And our system provides key frames included the face region for user as video summarized information.

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.5
• /
• pp.1-8
• /
• 2013

Smartphone enables diverse applications to be used in mobile environments. In spite of the high performance of smartphones, battery life has become one of the major constraints in mobility. Therefore, power efficiency of the smartphone is one of the most important factors in determining the efficiency of the smartphone. In this paper, in order to analyze the power efficiency of the smartphone, we have various experiments according to several configuration parameters such as processor, display and OS. We also use diverse applications. As a result, power consumption is dependent on the processor complexity and display size. However, power consumption shows the unpredictable pattern according to the OS. Smartphone using android OS consumes high power when internet and image processing applications are executed, but It consumes low power when music and camera applications are executed. In contrary, smartphone based on iOS consumes high power when game and internet applications are executed but it consumes low power when camera and processing applications are executed. In general, smartphone using iOS is more power efficient than smartphone based on android OS, because smartphone using iOS is optimized in the perspective of the hardware and OS.