• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.26 no.8B
• /
• pp.1129-1140
• /
• 2001

본 논문에서는 블록기반 동영상 부호화에서 나타나는 양자화 잡음을 그 특성에 맞게 모델링을 하고, 이를 기반으로 웨이블렛 변환(wavelet transform)을 이용하여 양자화 잡음을 제거하는 후처리 방법을 제안한다. 제안된 방법에서는 양자화 잡음을 특정 프로화일(profile)로 표현되는 블록화 잡음과 비에지 화소(non-edge pixel)에서 백색 가우시안 특성을 가지는 나머지 잡음의 합으로 모델링 한다. 이러한 양자화 잡음의 모델을 기반으로 정칙화 미분(regularized differentiation)을 표현하는 Mallat의 1차원 웨이브렛 변환을 이용하여 영상복원 관점에서 각각의 잡음을 제거한다. 먼저, 웨이브렛 영역의 블록경계에서 임펄스로 나타나는 블록화 잡음 성분들의 크기를 추정하여 줄임으로 해서 블록화 잡음을 제거한다. 이때 임펄스 크기의 추정은 메디안 필터와 양자화 파라미터(quantization parameter), 그리고 국부 활동도(local activity)를 이용하여 이루어진다. 그리고 나머지 잡음은 비에지 화소에서 연역치화(soft-thresholding)을 수행함으로써 제거한다. 이러한 후처리 방법의 구현은 실시간 응용을 위해 웨이브렛 필터를 이용하여 근사적으로 공간 영역에서 이루어진다. 실험 결과, 제안된 방법이 다양한 영상과 압축률에 대해 MPEG-4 VM(verification model) 후처리 필터(post-filter)보다 PSNR 성능뿐만 아니라 주관적 화질면에서도 우수함을 확인하였다.

• Journal of information and communication convergence engineering
• /
• v.10 no.3
• /
• pp.315-320
• /
• 2012

In H.264/AVC, the first frame of a group of pictures (GOP) is encoded in intra mode which generates a large number of bits. The number of bits for the I-frame affects the qualities of the following frames of a GOP since they are encoded using the bits remaining among the bits allocated to the GOP. In addition, the first frame is used for the inter mode encoding of the following frames. Thus, the initial quantization parameter (QP) affects the following frames as well as the first frame. In this paper, an adaptive peak signal to noise ratio (PSNR)-based initial QP determination algorithm is presented. In the proposed algorithm, a novel linear model is established based on the observation of the relation between the initial QPs and PSNRs of frames. Using the linear model and PSNR results of the encoded GOPs, the proposed algorithm accurately estimates the optimal initial QP which maximizes the PSNR of the current GOP. It is shown by experimental results that the proposed algorithm predicts the optimal initial QP accurately and thus achieves better PSNR performance than that of the existing algorithm.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.18 no.3
• /
• pp.89-100
• /
• 2023

Recently, on-device artificial intelligence (AI) solutions using mobile devices and embedded edge devices have emerged in various fields, such as computer vision, to address network traffic burdens, low-energy operations, and security problems. Although vision transformer deep learning models have outperformed conventional convolutional neural network (CNN) models in computer vision, they require more computations and parameters than CNN models. Thus, they are not directly applicable to embedded edge devices with limited hardware resources. Many researchers have proposed various model compression methods or lightweight architectures for vision transformers; however, there are only a few studies evaluating the effects of model compression techniques of vision transformers on performance. Regarding this problem, this paper presents a performance evaluation of vision transformers on embedded platforms. We investigated the behaviors of three vision transformers: DeiT, LeViT, and MobileViT. Each model performance was evaluated by accuracy and inference time on edge devices using the ImageNet dataset. We assessed the effects of the quantization method applied to the models on latency enhancement and accuracy degradation by profiling the proportion of response time occupied by major operations. In addition, we evaluated the performance of each model on GPU and EdgeTPU-based edge devices. In our experimental results, LeViT showed the best performance in CPU-based edge devices, and DeiT-small showed the highest performance improvement in GPU-based edge devices. In addition, only MobileViT models showed performance improvement on EdgeTPU. Summarizing the analysis results through profiling, the degree of performance improvement of each vision transformer model was highly dependent on the proportion of parts that could be optimized in the target edge device. In summary, to apply vision transformers to on-device AI solutions, either proper operation composition and optimizations specific to target edge devices must be considered.

• Convergence Security Journal
• /
• v.24 no.2
• /
• pp.9-17
• /
• 2024

This paper examines the security threats to enterprise Generative Artificial Intelligence systems and proposes countermeasures. As AI systems handle vast amounts of data to gain a competitive edge, security threats targeting AI systems are rapidly increasing. Since AI security threats have distinct characteristics compared to traditional human-oriented cybersecurity threats, establishing an AI-specific response system is urgent. This study analyzes the importance of AI system security, identifies key threat factors, and suggests technical and managerial countermeasures. Firstly, it proposes strengthening the security of IT infrastructure where AI systems operate and enhancing AI model robustness by utilizing defensive techniques such as adversarial learning and model quantization. Additionally, it presents an AI security system design that detects anomalies in AI query-response processes to identify insider threats. Furthermore, it emphasizes the establishment of change control and audit frameworks to prevent AI model leakage by adopting the cyber kill chain concept. As AI technology evolves rapidly, by focusing on AI model and data security, insider threat detection, and professional workforce development, companies can improve their digital competitiveness through secure and reliable AI utilization.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.24 no.8B
• /
• pp.1518-1531
• /
• 1999

The reconstructed images from highly compressed MPEG or H.263 data have noticeable image degradations, such as blocking artifacts near the block boundaries, corner outliers at cross points of blocks, and ringing noise near image edges, because the MPEG or H.263 quantizes the transformed coefficients of 8$\times$8 pixel blocks. A post-processing algorithm has been proposed by authors to reduce quantization effects, such as blocking artifacts, corner outliers, and ringing noise, in MPEG-decompressed images. Our signal-adaptive post-processing algorithm reduces the quantization effects adaptively by using both spatial frequency and temporal information extracted from the compressed data. The blocking artifacts are reduced by one-dimensional (1-D) horizontal and vertical low pass filtering (LPF), and the ringing noise is reduced by two-dimensional (2-D) signal-adaptive filtering (SAF). A comparison study of the subjective quality evaluation using modified single stimulus method (MSSM), the objective quality evaluation (PSNR) and the computation complexity analysis between the signal-adaptive post-processing algorithm and the MPEG-4 VM (Verification Model) post-processing algorithm is performed by computer simulation with several MPEG-4 image sequences. According to the comparison study, the subjective image qualities of both algorithms are similar, whereas the PSNR and the comparison complexity analysis of the signal-adaptive post-processing algorithm shows better performance than the VM post-processing algorithm.

• Journal of the Semiconductor & Display Technology
• /
• v.19 no.4
• /
• pp.71-76
• /
• 2020

In this paper, we propose a super-resolution method that reconstructs compressed low-resolution images into high-resolution images. We propose a CNN model with a small number of parameters, and even if quantization is applied to the proposed model, super-resolution can be implemented without deteriorating the image quality. To further improve the quality of the compressed low-resolution image, a new degradation model was proposed instead of the existing bicubic degradation model. The proposed degradation model is used only in the training process and can be applied by changing only the parameter values to the original CNN model. In the super-resolution image applying the proposed degradation model, visual artifacts caused by image compression were effectively removed. As a result, our proposed method generates higher PSNR values at compressed images and shows better visual quality, compared to conventional CNN-based SR methods.

• Journal of Broadcast Engineering
• /
• v.16 no.2
• /
• pp.216-225
• /
• 2011

Following the proliferation of three-dimensional video contents and displays, many terrestrial broadcasting companies have been preparing for stereoscopic 3DTV service. In terrestrial stereoscopic broadcast, it is a difficult task to code and transmit two video sequences while sustaining as high quality as 2DTV broadcast due to the limited bandwidth defined by the existing digital TV standards such as ATSC. Thus, a terrestrial 3DTV broadcasting with a heterogeneous video codec system, where the left image and right images are based on MPEG-2 and H.264/AVC, respectively, is considered in order to achieve both high quality broadcasting service and compatibility for the existing 2DTV viewers. Without significant change in the current terrestrial broadcasting systems, we propose a joint rate control scheme for stereoscopic 3DTV service based on the heterogeneous dual codec systems. The proposed joint rate control scheme applies to the MPEG-2 encoder a quadratic rate-quantization model which is adopted in the H.264/AVC. Then the controller is designed for the sum of the left and right bitstreams to meet the bandwidth requirement of broadcasting standards while the sum of image distortions is minimized by adjusting quantization parameter obtained from the proposed optimization scheme. Besides, we consider a condition on maintaining quality difference between the left and right images around a desired level in the optimization in order to mitigate negative effects on human visual system. Experimental results demonstrate that the proposed bit rate control scheme outperforms the rate control method where each video coding standard uses its own bit rate control algorithm independently in terms of the increase in PSNR by 2.02%, the decrease in the average absolute quality difference by 77.6% and the reduction in the variance of the quality difference by 74.38%.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.172-175
• /
• 2002

This paper presents perceptual model with a stochastic rnultiresolution characteristic that can be applied with watermark embedding in the biorthogonal wavelet domain. The perceptual model with adaptive watermarking algorithm embed at the texture and edge region for more strongly embedded watermark by the SSQ(successive subband quantization). The watermark embedding is based on the computation of a NVF(noise visibility function) that have local image properties. This method uses non-stationary Gaussian model stationary Generalized Gaussian model because watermark has noise properties. In order to determine the optimal NVF, we consider the watermark as noise. The particularities of embedding in the stationary GG model use shape parameter and variance of each subband regions in multiresolution. To estimate the shape parameter, we use a moment matching method. Non-stationary Gaussian model use the local mean and variance of each subband. The experiment results of simulation were found to be excellent invisibility and robustness. Experiments of such distortion are executed by Stirmark benchmark test.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.2E
• /
• pp.33-38
• /
• 1996

This paper provides an approach for representing an optimum vector quantizer by a scalar nonlinear gain-plus-additive noise model. The validity and accuracy of this analytic model is confirmed by comparing the calcuated model quantization errors with actual simulation of the optimum Linde-Buzo-Gray(LBG) vector quantizer. Using this model we frm MSE measure of an M-band filter bank codec in terms of the equivalent scalar quantizatin model and find the optimum FIR filter coefficients for each channel in the M-band structure for a given bit rate, given filter length, and given input signal correlation model. Specific design examples are worked out for 4-tap filters in the two-band paraunitary case. These theoretical results are confirmed by extensive Monte Carlo simulation.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2005.11a
• /
• pp.279-284
• /
• 2005

Knowledge based fault diagnosis has a limitation in determining the cause and scheme for the fault, because it detects faults from signal pattern only Therefore, model-based fault diagnosis is requested to determine the fault by analyzing output of the equipment from its dynamic model. This research shows a method how to devise the automaton of system as a model for normal and faulty condition through the reduction of handling data by quantization of vibration signals and the example which is concerning to the bearing of ATM. The developed model based fault diagnosis was applied to detect the faulty bearing of ATM, which results.