Multi-layer and Multi-scale feature aggregation for DIBR-Synthesized image quality assessment

Depth-Image-Based Rendering (DIBR) is one of the main fundamental techniques for generating new viewpoints in 3D video applications such as multi-viewpoint video (MVV), free viewpoint video (FVV) and virtual reality (VR). Due to the imperfections of color images, depth maps or texture restoration techniques, several types of distortions occur in synthesized views. However, most of related works evaluated the quality of DIBR-synthesized views by only detecting a specific type of distortion, such as stretching, black holes, blurring, etc., which were unable to accurately evaluate the quality of DIBR-synthesized views. In this paper, a new no-reference image quality assessment method is proposed to evaluate the quality of DIBR-synthesized images by combining multi-layer and multi-scale features of images. To be specific, the distortions introduced by different stages of virtual viewpoint synthesis are first analyzed, and then multi-layer and multi-scale features are extracted to estimate the degree of texture and structure distortions. As a result, individual quality scores associated with two types of distortions (e.g., structural distortion and texture distortion) are aggregated to an overall image quality. Experimental results on two publicly available DIBR datasets show that the method has better performance than the state-of-the-art models.Index Terms: image quality assessment, DIBR-synthesized image, distortion correction, BIQA.     

Multi-objective routing algorithms for low-earth orbit satellite network

The low-earth orbit (LEO) satellite network, composed of a large number of satellite nodes, is a hot research topic at present. Due to the characteristics of the large-scale LEO satellite network, such as many satellite nodes, short orbit period, large dynamic change of topology, and unstable link-state, its communication quality of service (QoS) requirements are difficult to meet. Aiming at this problem, various factors that may affect data transmission are first analyzed. The network link selection problem is modeled as a multi-constraint optimization decision problem, a routing mathematical model based on linear programming (LP) is designed, and its solution is solved. Aiming at the problem of limited onboard computing resources, a multi-object optimization Dijkstra algorithm (MOODA) is designed. The MOODA finds the optimal path according to the comprehensive performance of the link. It solves the problems of poor comprehensive QoS performance and the low degree of load balancing of the paths found by the Dijkstra algorithm. The simulation results show that the paths found by the two algorithms have good QoS, robustness, and load balancing performance.     

Edge Computing Platform with Efficient Migration Scheme for 5G/6G Networks

Next-generation cellular networks are expected to provide users with innovative gigabits and terabits per second speeds and achieve ultra-high reliability, availability, and ultra-low latency. The requirements of such networks are the main challenges that can be handled using a range of recent technologies, including multi-access edge computing (MEC), artificial intelligence (AI), millimeter-wave communications (mmWave), and software-defined networking. Many aspects and design challenges associated with the MEC-based 5G/6G networks should be solved to ensure the required quality of service (QoS). This article considers developing a complex MEC structure for fifth and sixth-generation (5G/6G) cellular networks. Furthermore, we propose a seamless migration technique for complex edge computing structures. The developed migration scheme enables services to adapt to the required load on the radio channels. The proposed algorithm is analyzed for various use cases, and a test bench has been developed to emulate the operator’s infrastructure. The obtained results are introduced and discussed.     

Topology Driven Cooperative Self Scheduling for Improved Lifetime Maximization in WSN

In Wireless Sensor Network (WSN), scheduling is one of the important issues that impacts the lifetime of entire WSN. Various scheduling schemes have been proposed earlier to increase the lifetime of the network. Still, the results from such methods are compromised in terms of achieving high lifetime. With this objective to increase the lifetime of network, an Efficient Topology driven Cooperative Self-Scheduling (TDCSS) model is recommended in this study. Instead of scheduling the network nodes in a centralized manner, a combined approach is proposed. Based on the situation, the proposed TDCSS approach performs scheduling in both the ways. By sharing the node statistics in a periodic manner, the overhead during the transmission of control packets gets reduced. This in turn impacts the lifetime of all the nodes. Further, this also reduces the number of idle conditions of each sensor node which is required for every cycle. The proposed method enables every sensor to schedule its own conditions according to duty cycle and topology constraints. Central scheduler monitors the network conditions whereas total transmissions occurs at every cycle. According to this, the source can infer the possible routes in a cycle and approximate the available routes. Further, based on the statistics of previous transmissions, the routes towards the sink are identified. Among the routes found, a single optimal route with energy efficiency is selected to perform data transmission. This cooperative approach improves the lifetime of entire network with high throughput performance.     

基于多视角的多类型错误全面检测方法

随着信息化社会的发展,数据的规模越发庞大,数据的种类也越发丰富.时至今日,数据已经成为国家和企业的重要战略资源,是科学化管理的重要保障.然而,随着社会生活产生的数据日益丰富,大量的脏数据也随之而来,数据质量问题油然而生.如何准确而全面地检测出数据集中所包含的错误数据,一直是数据科学中的痛点问题.尽管已有许多传统方法被广泛用于各行各业,如基于约束与统计的检测方法,但这些方法通常需要丰富的先验知识与昂贵的人力和时间成本.受限于此,这些方法往往难以准确而全面地检测数据.近年来,许多新型错误检测方法利用深度学习技术,通过时序推断、文本解析等方式取得了更好检测效果,但它们通常只适用于特定的领域或特定的错误类型,面对现实生活中的复杂情况,泛用性不足.基于上述情况,结合传统方法与深度学习技术的优点,提出了一个基于多视角的多类型错误全面检测模型CEDM.首先,从模式的角度,结合现有约束条件,在属性、单元和元组层面进行多维度的统计分析,构建出基础检测规则;然后,通过词嵌入捕获数据语义,从语义的角度分析属性相关性、单元关联性与元组相似性,进而基于语义关系,从多个维度上更新、扩展基础规则;最终,联合多个视角...     

软件缺陷预测模型间的比较实验: 问题、进展与挑战

近年来,研究者提出了大量的软件缺陷预测模型,新模型往往通过与过往模型进行比较实验来表明其有效性.然而,研究者在设计新旧模型间的比较实验时并没有达成共识,不同的工作往往采用不完全一致的比较实验设置,这可能致使在对比模型时得到误导性结论,最终错失提升缺陷预测能力的机会.对近年来国内外学者所做的缺陷预测模型间的比较实验进行系统性的总结:首先,阐述缺陷预测模型间的比较实验的研究问题;然后,分别从缺陷数据集、数据集划分、基线模型、性能指标、分类阈值这5个方面对现有的比较实验进行总结;最后,指出目前在进行缺陷预测模型间比较实验时面临的挑战,并给出建议的研究方向.     

深度学习点云质量增强方法综述

随着三维探测技术的发展,点云逐渐成为最常用的三维物体/场景表征数据类型之一,广泛应用于自动驾驶、增强现实及虚拟现实等领域。然而,受限于硬件设备、采集环境以及遮挡等因素,采集的原始点云通常是不完整、稀疏、嘈杂的,为点云的处理和分析带来了巨大挑战。在此背景下,点云质量增强技术旨在对原始点云进行处理以获得结构完整、密集且接近无噪的点云,具有重要意义。本文对现阶段深度学习点云质量增强方法进行了系统综述,为后续研究者提供研究基础。首先,简要介绍了点云数据处理中通用的关键技术;分别介绍了补全、上采样和去噪3类点云质量增强方法,并对3类方法中的现有算法进行了分类、梳理及总结。其中,点云补全与点云去噪算法均可根据是否采用编码器—解码器结构分为两大类,点云上采样算法可根据网络主要结构分为基于卷积神经网络的方法、基于生成对抗网络的方法和基于图卷积神经网络的方法。其次,总结了质量增强任务中常用的数据集与评价指标,并分别对比分析了现阶段点云补全、上采样和去噪中主流算法的性能。最后,通过系统的梳理,凝练出点云质量增强方向所面临的挑战,并对未来的研究趋势进行了展望。此外,本文汇总了涉及的文献及其开源代码,详见链接https：//github.com/LilydotEE/Point_cloud_quality_enhancement。     

深度学习时代图像融合技术进展

为提升真实场景视觉信号的采集质量,往往需要通过多种融合方式获取相应的图像,例如,多聚焦、多曝光、多光谱和多模态等。针对视觉信号采集的以上特性,图像融合技术旨在利用同一场景不同视觉信号的优势,生成单图像信息描述,提升视觉低、中、高级任务的性能。目前,依托端对端学习强大的特征提取、表征及重构能力,深度学习已成为图像融合研究的主流技术。与传统图像融合技术相比,基于深度学习的图像融合模型性能显著提高。随着深度学习研究的深入,一些新颖的理论和方法也促进了图像融合技术的发展,如生成对抗网络、注意力机制、视觉Transformer和感知损失函数等。为厘清基于深度学习技术的图像融合研究进展,本文首先介绍了图像融合问题建模,并从传统方法视角逐渐向深度学习视角过渡。具体地,从数据集生成、神经网络构造、损失函数设计、模型优化和性能评估等方面总结了基于深度学习的图像融合研究现状。此外,还讨论了选择性图像融合这类衍生问题建模(如基于高分辨率纹理图融合的深度图增强),回顾了一些基于图像融合实现其他视觉任务的代表性工作。最后,根据现有技术的缺陷,提出目前图像融合技术存在的挑战,并对未来发展趋势给出了展望。     

考虑双目竞争视觉现象的非对称失真立体图像质量评价方法

目的 现有方法存在特征提取时间过长、非对称失真图像预测准确性不高的问题,同时少有工作对非对称失真与对称失真立体图像的分类进行研究,为此提出了基于双目竞争的非对称失真立体图像质量评价方法。方法 依据双目竞争的视觉现象,利用非对称失真立体图像两个视点的图像质量衰减程度的不同,生成单目图像特征的融合系数,融合从左右视点图像中提取的灰度空间特征与HSV （hue-saturation-value）彩色空间特征。同时,量化两个视点图像在结构、信息量和质量衰减程度等多方面的差异,获得双目差异特征。并且将双目融合特征与双目差异特征级联为一个描述能力更强的立体图像质量感知特征向量,训练基于支持向量回归的特征—质量映射模型。此外,还利用双目差异特征训练基于支持向量分类模型的对称失真与非对称失真立体图像分类模型。结果 本文提出的质量预测模型在4个数据库上的SROCC （Spearman rank order correlation coefficient）和PLCC （Pearson linear correlation coefficient）均达到0.95以上,在3个非对称失真数据库上的均方根误差（root of mean square error,RMSE）取值均优于对比算法。在LIVE-II（LIVE 3D image quality database phase II）、IVC-I（Waterloo-IVC 3D image qualityassessment database phase I）和IVC-II （Waterloo-IVC 3D image quality assessment database phase II）这3个非对称失真立体图像测试数据库上的失真类型分类测试中,对称失真立体图像的分类准确率分别为89.91%、94.76%和98.97%,非对称失真立体图像的分类准确率分别为95.46%,92.64%和96.22%。结论 本文方法依据双目竞争的视觉现象融合左右视点图像的质量感知特征用于立体图像质量预测,能够提升非对称失真立体图像的评价准确性和鲁棒性。所提取双目差异性特征还能够用于将对称失真与非对称失真立体图像进行有效分类,分类准确性高。     

基于多特征融合的无监督真值发现方法

真值发现是数据集成领域具有挑战性的研究热点之一。传统的方法利用数据源与观测值之间的交互关系推断真值,缺乏足够的特征信息;基于深度学习的方法可以有效地进行特征抽取,但其性能依赖于大量手工标注,而在实际应用中很难获取到大量高质量的真值标签。为克服以上问题,本文提出一种基于多特征融合的无监督真值发现方法（Unsupervised truth discovery method based on multi-feature fusion, MFOTD）。首先,利用集成学习无监督标注“真值”标签;然后,分别使用预训练模型 Bert和独热编码获取观测值的语义特征和交互特征;最后,融合观测值多种特征并使用其“真值”标签构建初始训练集,通过自训练方式训练真值预测模型。在两个真实数据集上的实验结果表明,与已有方法相比,本文所提出的方法具有更高的真值发现准确性。