We propose a context-adaptive entropy model for use in end-to-end optimized image compression. Our exploits two types of contexts, bit-consuming contexts and bit-free distinguished based upon whether additional bit allocation is required. Based on these we allow the to more accurately estimate distribution each latent representation with generalized form approximation models, which accordingly leads an enhanced compression performance. experimental results, proposed method outperforms...

Content-based image retrieval has been actively studied in several fields. This provides more effective management and of images than the keyword-based approach. However, most conventional methods lack capability to effectively incorporate human intuition emotion into retrieving images. It is difficult obtain satisfactory results when user wants that cannot be explicitly described or can requested only based on impression. In order solve this problem supplement user's expression capability,...

Automatic analysis of the video is one most complex problems in fields computer vision and machine learning. A significant part this research deals with (human) activity recognition (HAR) since humans, activities that they perform, generate semantics. Video-based HAR has applications various domains, but important challenging sports videos. Some major issues include high inter- intra-class variations, large class imbalance, presence both group actions single player actions, recognizing...

Recently, learned image compression methods have been actively studied. Among them, entropy-minimization based approaches achieved superior results compared to conventional codecs such as BPG and JPEG2000. However, the quality enhancement rate-minimization are conflictively coupled in process of compression. That is, maintaining high entails less vice versa. by jointly training separate conjunction with compression, coding efficiency can be improved. In this paper, we propose a novel joint...

Abstract Recently, learned image compression methods based on entropy minimization have achieved superior results compared with conventional codecs such as BPG and JPEG2000. However, they leverage single Gaussian models, which a limited ability to approximate various irregular distributions of transformed latent representations, resulting in suboptimal coding efficiency. Furthermore, existing focus constructing effective rather than utilizing modern architectural techniques. In this paper,...

The proliferation of deep learning-based machine vision applications has given rise to a new type compression, so called video coding for (VCM). VCM differs from traditional in that it is optimized performance instead human visual quality. In the feature compression track MPEG-VCM, multi-scale features extracted images are subject compression. Recent works have demonstrated versatile (VVC) standard-based approach can achieve BD-rate reduction up 96% against MPEG-VCM anchor. However, still...

The component of cell parasitic resistance at sub-20nm 4th generation DRAM transistor is investigated. To evaluate the characteristics, Gate Buried Contact (GBC) to Active contact formation method with varied dopant concentrations was studied. We have discovered a scalable methodology that simultaneously reduces and leakage regard Induced Drain Leakage (GIDL). Also, we proved importance interface quality Direct on Cell (DCC) in order reduce resistance. failure analysis conducted by...

In this paper, a low bit-rate compressed image quality enhancement framework is presented. A recent image/video coding method and deep learning based are integrated to improve the perceptual of images. The proposed architecture designed reduce artifact restore blurred texture details. experimental results presents that yields 33% improvement in Perceptual Index score which consistent with visual evaluation on sample results.

Machine vision-based intelligent applications that analyze video data collected by machines are rapidly increasing. Therefore, it is essential to efficiently compress a large volume of for machine consumption. Accordingly, the Moving Picture Experts Group (MPEG) has been developing new coding standard called Video Coding Machines (VCM), aimed at consumed rather than humans. Recently, studies have demonstrated multi-scale feature compression (MSFC)-based methods significantly improve...

Recently, "Speed" is one of the hot issues in digital forensics. Thanks to a recent advanced technology, today we can get bigger hard drive disks at lower price than previously. But unfortunately, it means for forensic investigators that they need tremendous time and effort sequence process creating images, searching into them analyzing them. In order solve this problem, some methods have been proposed improve performance tools. One getting attention hardware-based approach. However, such...

Recently, neural-network based lossy image compression methods have been actively studied and they achieved remarkable performance. However, the classical evaluation metrics, such as PSNR MS-SSIM, that recent approaches using in their objective function yield sub-optimal coding efficiency terms of human perception, although are very dominant metrics research standardization fields. Taking into account improving perceptual quality is one major goals compression, we propose a new training...

Recently, many neural network-based image compression methods have shown promising results superior to the existing tool-based conventional codecs. However, most of them are often trained as separate models for different target bit rates, thus increasing model complexity. Therefore, several studies been conducted learned that supports variable rates with single models, but they require additional network modules, layers, or inputs lead complexity overhead, do not provide sufficient coding...

In this paper, we demonstrate the structure and characteristics of bandwidth-efficient stereoscopic 3-D broadcasting system entitled service compatible 3-D-TV using main mobile hybrid delivery (SC-MMH), which enables broadcasters to provide three different types services (2-D fixed, mobile, 3-D) by transmitting only two legacy broadcast streams fixed mobile). By combining existing networks, proposed provides high quality services, fully guarantees backward-compatibility with fixed/mobile 2-D...

The emerging versatile video coding (VVC) standard currently adopts 67 intra-prediction modes in order to improve the performance. most probable mode (MPM) is used encode prediction efficiently based on of neighbouring blocks. Due an increase number intra-modes and resolution input sequence, it necessary intra-mode current block. This Letter proposes efficient method extending MPM called frequent (MFM), which exploits occurrences proposed MFM derives intra-mode. Then derived signaled by a...

Packet recording or capturing is one of the most useful tools for network forensics and surveillance. Since a storage system limited size, de-duplication can be used to save disk space. In this article, we present new scalable engine packet that eliminate redundant contents over multiple packets. Unlike previous work, our proposed scheme designed packet-level support any kinds from current Internet emerging networks. We also fast chunking method indexing enable instances execute in parallel....

Recently, neural network (NN)-based image compression studies have actively been made and has shown impressive performance in comparison to traditional methods. However, most of the works focused on non-scalable (single-layer coding) while spatially scalable drawn less attention although it many applications. In this paper, we propose a novel NN-based method, called COMPASS, which supports arbitrary-scale spatial scalability. Our proposed COMPASS very flexible structure where number layers...

In search for efficient feature compression technologies machine consumption, MPEG recently issued a call proposal (CfP) on video coding (FCVCM). One issue in is that the input maps generally have high redundancy them. Various researches to reduce such been made. For example, recent study called L-MSFC (learnable multi-scale compression), which effectively combines fusion and an end-to-end learnable framework, showed up 98% BD rate gain over anchor model defined FCVCM CfP. Despite these...

As cyber attacks have increased in recent years, network forensics, which collects and analyses packets as well digital has been studied. However, high-speed networks such 1 or 10 Gbps many flows. For example, a hundreds of millions flows per day. Analysing traffic this situation is very difficult time-consuming. In paper, we propose system that can analyse abnormal behaviour quickly easily. We first stores the TCP flag when generating Second, present some ways to use anomalies persistent...