To compress screen image sequence in real-time remote and interactive applications,a novel compression method is proposed.The proposed method is named as CABHG.CABHG employs hybrid coding schemes that consist of intra...To compress screen image sequence in real-time remote and interactive applications,a novel compression method is proposed.The proposed method is named as CABHG.CABHG employs hybrid coding schemes that consist of intra-frame and inter-frame coding modes.The intra-frame coding is a rate-distortion optimized adaptive block size that can be also used for the compression of a single screen image.The inter-frame coding utilizes hierarchical group of pictures(GOP) structure to improve system performance during random accesses and fast-backward scans.Experimental results demonstrate that the proposed CABHG method has approximately 47%-48% higher compression ratio and 46%-53% lower CPU utilization than professional screen image sequence codecs such as TechSmith Ensharpen codec and Sorenson 3 codec.Compared with general video codecs such as H.264 codec,XviD MPEG-4 codec and Apple's Animation codec,CABHG also shows 87%-88% higher compression ratio and 64%-81% lower CPU utilization than these general video codecs.展开更多
In this paper, an investigation is made on compatibility of the video formats TV (625/50/2:1/4:3), EDTV (625/50/1:1/16:9), HDTVint (1250/50/2:1/16:9) and HDTVpro(1250/50/2:1/16:9) in three-dimensional subband coding. ...In this paper, an investigation is made on compatibility of the video formats TV (625/50/2:1/4:3), EDTV (625/50/1:1/16:9), HDTVint (1250/50/2:1/16:9) and HDTVpro(1250/50/2:1/16:9) in three-dimensional subband coding. A hierarchical method is proposed leading to a more reasonable joint solution for the issues of digital transmission evolution and interlace-to-progressive scanning conversion ill the frame work of bit rate compression. Conventional HDTVint with the diamond shaped spectrum is proposed to be replaced by a kind of 'switching'HDTVin signal which consists of three components and motion detection bits (MDB). The first component and the MDB represent TV. The first plus the second component and the MDB represent EDTV.HDTVpro is splitted into switching HDTVint and extra component. The switching threshold and the length of temporal filter banks are discussed. Rearranging the pixels in subbands is recommended. Based on the proposed method, the transmission system can evolve through four steps: TV, EDTV, HDTVint and HDTVpro Results of bit rate compression are also presented. TV, EDTV, HDTVint and HDTVpro are compressed to approximately 40, 80, 128 and 256 Mbit/s respectively. Interlace scanning has drawbacks such as line flicker and line crawling as the eye follows the line structure in the picture. Progressive version switching TV and HDTVint can be reconstructed using one frame memory without the problems of handling temporal-vertical diamond shaped spectra of conventional TV and HDTVint based on a one-frame memory algorithm in the receiver.展开更多
当前基于深度学习的中文长文本摘要生成的研究存在以下问题:(1)生成模型缺少信息引导,缺乏对关键词汇和语句的关注,存在长文本跨度下关键信息丢失的问题;(2)现有中文长文本摘要模型的词表常以字为基础,并不包含中文常用词语与标点,不利...当前基于深度学习的中文长文本摘要生成的研究存在以下问题:(1)生成模型缺少信息引导,缺乏对关键词汇和语句的关注,存在长文本跨度下关键信息丢失的问题;(2)现有中文长文本摘要模型的词表常以字为基础,并不包含中文常用词语与标点,不利于提取多粒度的语义信息.针对上述问题,本文提出了融合引导注意力的中文长文本摘要生成(Chinese Long text Summarization with Guided Attention,CLSGA)方法.首先,针对中文长文本摘要生成任务,利用抽取模型灵活抽取长文本中的核心词汇和语句,构建引导文本,用以指导生成模型在编码过程中将注意力集中于更重要的信息.其次,设计中文长文本词表,将文本结构长度由字统计改变至词组统计,有利于提取更加丰富的多粒度特征,进一步引入层次位置分解编码,高效扩展长文本的位置编码,加速网络收敛.最后,以局部注意力机制为骨干,同时结合引导注意力机制,以此有效捕捉长文本跨度下的重要信息,提高摘要生成的精度.在四个不同长度的公共中文摘要数据集LCSTS(大规模中文短文本摘要数据集)、CNewSum(大规模中国新闻摘要数据集)、NLPCC2017和SFZY2020上的实验结果表明:本文方法对于长文本摘要生成具有显著优势,能够有效提高ROUGE-1、ROUGE-2、ROUGE-L值.展开更多
随着纠删码在分布式存储系统中的实际应用,纠删码为存储系统提供了更加优秀的存储效率,但当节点丢失时,相较于传统副本技术更多的网络传输带宽开销成为了造成系统性能瓶颈的关键因素。为了解决MDS编码高带宽开销对系统性能的影响,一类...随着纠删码在分布式存储系统中的实际应用,纠删码为存储系统提供了更加优秀的存储效率,但当节点丢失时,相较于传统副本技术更多的网络传输带宽开销成为了造成系统性能瓶颈的关键因素。为了解决MDS编码高带宽开销对系统性能的影响,一类新型编码方案——分组码被应用在分布式存储系统中,相较于传统MDS编码能够有效地降低节点修复时的数据传输量,从而减少网络带宽需求。在Pyramid分组码的基础上进行层次扩展,提出一种HLRC(hierarchical local repair codes)纠删码。HLRC相较于LRC引入了层次编码模型,将原始数据块构建为编码矩阵,根据层次进行分别编码,生成包含数据块范围不同的局部校验块;每个层次包含的数据块数量不同,可以保证修复节点时的低修复成本,同时还拥有较高的存储效率。HLRC相较于Pyramid拥有额外的校验块冗余,能够降低校验块出错和多节点出错时的恢复开销。在基于Ceph的分布式存储系统中的实验结果表明,HLRC与Pyramid等分组码相比,单节点修复开销最高可降低48.56%,多节点修复开销最高可降低25%。展开更多
高效的图像特征表示是计算机视觉的基础.基于图像的视觉显著性机制及深度学习模型的思想,提出一种融合图像显著性的层次稀疏特征表示用于图像分类.这种层次特征学习每一层都由3个部分组成:稀疏编码、显著性最大值汇聚(saliency max pool...高效的图像特征表示是计算机视觉的基础.基于图像的视觉显著性机制及深度学习模型的思想,提出一种融合图像显著性的层次稀疏特征表示用于图像分类.这种层次特征学习每一层都由3个部分组成:稀疏编码、显著性最大值汇聚(saliency max pooling)和对比度归一化.通过在图像层次稀疏表示中引入图像显著信息,加强了图像特征的语义信息,得到图像显著特征表示.相比于手工指定特征,该模型采用无监督数据驱动的方式直接从图像中学习到有效的图像特征描述.最后采用支持向量机(support vector machine,SVM)分类器进行监督学习,实现对图像进行分类.在2个常用的标准图像数据集(Caltech 101和Caltech 256)上进行的实验结果表明,结合图像显著性信息的层次特征表示,相比于基于局部特征的单层稀疏表示在分类性能上有了显著提升.展开更多
基金Project(60873230) supported by the National Natural Science Foundation of China
文摘To compress screen image sequence in real-time remote and interactive applications,a novel compression method is proposed.The proposed method is named as CABHG.CABHG employs hybrid coding schemes that consist of intra-frame and inter-frame coding modes.The intra-frame coding is a rate-distortion optimized adaptive block size that can be also used for the compression of a single screen image.The inter-frame coding utilizes hierarchical group of pictures(GOP) structure to improve system performance during random accesses and fast-backward scans.Experimental results demonstrate that the proposed CABHG method has approximately 47%-48% higher compression ratio and 46%-53% lower CPU utilization than professional screen image sequence codecs such as TechSmith Ensharpen codec and Sorenson 3 codec.Compared with general video codecs such as H.264 codec,XviD MPEG-4 codec and Apple's Animation codec,CABHG also shows 87%-88% higher compression ratio and 64%-81% lower CPU utilization than these general video codecs.
文摘In this paper, an investigation is made on compatibility of the video formats TV (625/50/2:1/4:3), EDTV (625/50/1:1/16:9), HDTVint (1250/50/2:1/16:9) and HDTVpro(1250/50/2:1/16:9) in three-dimensional subband coding. A hierarchical method is proposed leading to a more reasonable joint solution for the issues of digital transmission evolution and interlace-to-progressive scanning conversion ill the frame work of bit rate compression. Conventional HDTVint with the diamond shaped spectrum is proposed to be replaced by a kind of 'switching'HDTVin signal which consists of three components and motion detection bits (MDB). The first component and the MDB represent TV. The first plus the second component and the MDB represent EDTV.HDTVpro is splitted into switching HDTVint and extra component. The switching threshold and the length of temporal filter banks are discussed. Rearranging the pixels in subbands is recommended. Based on the proposed method, the transmission system can evolve through four steps: TV, EDTV, HDTVint and HDTVpro Results of bit rate compression are also presented. TV, EDTV, HDTVint and HDTVpro are compressed to approximately 40, 80, 128 and 256 Mbit/s respectively. Interlace scanning has drawbacks such as line flicker and line crawling as the eye follows the line structure in the picture. Progressive version switching TV and HDTVint can be reconstructed using one frame memory without the problems of handling temporal-vertical diamond shaped spectra of conventional TV and HDTVint based on a one-frame memory algorithm in the receiver.
文摘当前基于深度学习的中文长文本摘要生成的研究存在以下问题:(1)生成模型缺少信息引导,缺乏对关键词汇和语句的关注,存在长文本跨度下关键信息丢失的问题;(2)现有中文长文本摘要模型的词表常以字为基础,并不包含中文常用词语与标点,不利于提取多粒度的语义信息.针对上述问题,本文提出了融合引导注意力的中文长文本摘要生成(Chinese Long text Summarization with Guided Attention,CLSGA)方法.首先,针对中文长文本摘要生成任务,利用抽取模型灵活抽取长文本中的核心词汇和语句,构建引导文本,用以指导生成模型在编码过程中将注意力集中于更重要的信息.其次,设计中文长文本词表,将文本结构长度由字统计改变至词组统计,有利于提取更加丰富的多粒度特征,进一步引入层次位置分解编码,高效扩展长文本的位置编码,加速网络收敛.最后,以局部注意力机制为骨干,同时结合引导注意力机制,以此有效捕捉长文本跨度下的重要信息,提高摘要生成的精度.在四个不同长度的公共中文摘要数据集LCSTS(大规模中文短文本摘要数据集)、CNewSum(大规模中国新闻摘要数据集)、NLPCC2017和SFZY2020上的实验结果表明:本文方法对于长文本摘要生成具有显著优势,能够有效提高ROUGE-1、ROUGE-2、ROUGE-L值.
文摘随着纠删码在分布式存储系统中的实际应用,纠删码为存储系统提供了更加优秀的存储效率,但当节点丢失时,相较于传统副本技术更多的网络传输带宽开销成为了造成系统性能瓶颈的关键因素。为了解决MDS编码高带宽开销对系统性能的影响,一类新型编码方案——分组码被应用在分布式存储系统中,相较于传统MDS编码能够有效地降低节点修复时的数据传输量,从而减少网络带宽需求。在Pyramid分组码的基础上进行层次扩展,提出一种HLRC(hierarchical local repair codes)纠删码。HLRC相较于LRC引入了层次编码模型,将原始数据块构建为编码矩阵,根据层次进行分别编码,生成包含数据块范围不同的局部校验块;每个层次包含的数据块数量不同,可以保证修复节点时的低修复成本,同时还拥有较高的存储效率。HLRC相较于Pyramid拥有额外的校验块冗余,能够降低校验块出错和多节点出错时的恢复开销。在基于Ceph的分布式存储系统中的实验结果表明,HLRC与Pyramid等分组码相比,单节点修复开销最高可降低48.56%,多节点修复开销最高可降低25%。
