基于多模态推理图神经网络的场景文本视觉问答模型

文本阅读能力差和视觉推理能力不足是现有视觉问答(visual question answering, VQA)模型效果不好的主要原因,针对以上问题,设计了一个基于图神经网络的多模态推理(multi-modal reasoning graph neural network, MRGNN)模型。利用图像中多种形式的信息帮助理解场景文本内容,将场景文本图片分别预处理成视觉对象图和文本图的形式,并且在问题自注意力模块下过滤多余的信息;使用加入注意力的聚合器完善子图之间相互的节点特征,从而融合不同模态之间的信息,更新后的节点利用不同模态的上下文信息为答疑模块提供了更好的功能。在ST-VQA和TextVQA数据集上验证了有效性,实验结果表明,相比较此任务的一些其他模型,MRGNN模型在此任务上有明显的提升。     

Evaluation of contrast metrics for liquid‐crystal displays under different viewing conditions

Abstract— A psychophysical experiment was carried out to assess the perceptual contrast on a large‐sized liquid‐crystal display (LCD) under nine phases of viewing conditions. Based on the results, six contrast models employing different color attributes or the index of just‐noticeable difference (JND) were developed. Their performances were also tested by the visual data of the nine phases, which indicated that the model C_Q employing CIECAM02 brightness gives the best performance in predicting visual data under different viewing conditions among those models. A preprocessing step was also proposed to utilize the contrast model C_Q conveniently in the practical contrast evaluation of LCDs.     

Web模式的仿真想定编辑与管理系统设计

基于虚拟战场对战场态势信息的需求,构建了Web模式的仿真想定编辑与管理系统,该系统包括仿真想定的输入,管理,编辑,标绘,输出等功能模块。在地理信息系统（Geographic Information System,GIS）和军标库等的支持下,用户可以根据具体的仿真需求,通过对仿真想定进行可视化编辑,简单灵活地实现对战场初始态势信息的输入。同时系统加强对已有的仿真想定的统一管理,建立仿真想定库,便于用户获取已有的仿真想定,大大提高其重用率,减少了不必要的重复劳动。     

Weber定律下尺度空间的自适应构建

在Weber定律的启发下提出一种基于"恰可识别差异"的尺度空间自适应构建方法。Weber定律认为人类感知模式不仅与刺激的变化有关,而且与原刺激强度有关。基于此观点,在图像尺度空间的构建过程中,首先以Marr视觉理论的屋脊型边缘和阶梯型边缘的特征之和计算图像的信息量,再通过实验获得人类视觉能感知到"恰可识别差异"时的图像信息量变化,最后通过曲线拟合自适应构建图像的尺度空间。实验结果表明,该方法充分体现了人类视觉感知特性,与其他尺度空间构建方法相比,在目标匹配实验中,匹配数目提升了25%以上;在去噪实验中,本算法的去噪效果良好。     

小波域最小嵌入失真函数设计及其在隐写中的应用

为了提高隐写方案的安全性,提出一种基于最小嵌入失真原理和网格码的图像隐写算法。首先在离散小波域结合人眼视觉特性和整数提升小波变换设计了失真测度函数,主要考虑了亮度、频率和纹理掩蔽因子对载体失真的影响。然后结合网格码设计了隐写算法,将嵌入信息对载体的修改最小化并且集中在人眼不敏感区域。实验结果显示,方案具有良好的视觉不可见性,且能抵抗空域、小波域等隐写分析的攻击,安全容量达到0.4 bits/pixel。     

应用改进频率调谐的海上小目标检测方法

提出了一种新的基于可见光图像的海上小目标检测方法。由于频率调谐方法将图像空间域整体均值与高斯滤波后差分结果,作为显著性度量的标准,因此当图像背景中存在较多杂波干扰时,显著目标检测效果不理想。提出的方法对基于频率调谐的显著性检测方法进行了改进,首先对图像LAB空间中3个特征分量进行分块,在每个分块区域中应用频率调谐显著性检测方法,进而将其结果合并为总显著图,以检测海上小目标。该方法克服了频率调谐方法,当海面背景中存在大量海杂波,无法有效提取小目标的缺陷。实验结果表明了该方法的有效性。     

基于感知驱动的纹饰矢量图评价方法

针对构建中华文化矢量素材库过程中如何评价矢量图质量优劣的问题,提出精确性、连续性和简单性三大视觉评价指标,并进行了量化分析. 首先,对矢量图进行解析,获取矢量图的文档对象模型,遍历文档树得到矢量信息;然后,结合格式塔心理学对矢量图的视觉误差、曲线连续性和纹饰图案的复杂度进行量化;最后,提出一种基于感知驱动的传统纹饰矢量图的评价方法. 结合视觉上的启发性规则,所提评价方法可提供3个量化指标用于多任务驱动矢量化系统中矢量图的自动评价、排序和筛选. 在纹饰矢量图数据集上对算法进行了仿真实验,结果表明,对矢量图的评价结果符合主流设计的视觉审美和期望.     

两相闭式热虹吸管冷凝换热规律的实验研究

本文通过可视化热虹吸管实验,拍摄了冷凝段内部冷凝液的各种基本流型和特殊流型,设计了试验用多节风冷热虹吸管以及相应的风道系统。采用自行组装的高精度微机控制的测试系统测量温度,对不同工质(水、乙醇、丙酮、氟里昂-11)热虹吸管进行了不同工况的实验。获得了各种工程实际工况下的冷凝换热系数,还获得了冷凝换热系数随热流密度、蒸汽压力变化的规律以及沿2.5米长冷凝面长度上的变化规律。实验结果表明,热虹吸管的冷凝换热系数在很大范围内偏离努塞尔特理论解。本文结合可视化实验,对其进行了分析。     

Measuring individual corrective reaction time using the intermittent illumination model

The corrective reaction time (t_cr) is an essential motor property when modelling hand control movements. Many studies designed experiments to estimate t_cr, but reported only group means with inconsistent definitions. This study proposes an alternative methodology using Drury's (1994) intermittent illumination model. A total of 24 participants performed circular tracking movements under five levels of visual information delay using a modified monitor in a darkened room. Measured movement speeds and the manipulated delays were used with the model to estimate t_cr of individuals and test effects of gender and path width. The results showed excellent model fits and demonstrated individual differences of t_cr, which was 273 ms on average and ranged from 87 to 441 ms. The wide range of t_cr values was due to significant effects of gender and path width. Male participants required shorter t_cr compared to female participants, especially for narrow path widths.     

Utility of stereoscopic displays for indirect-vision driving and robot teleoperation

The effectiveness of an active shutter-glasses stereoscopic display (SD) and a passive polarised SD was evaluated in a live robot-teleoperation task and a simulated indirect-vision driving task in various terrains. Overall, participants completed their tasks significantly faster with the SDs in three-dimensional (3D) mode than with the SDs in the baseline 2D mode. They also navigated more accurately with the SDs in 3D mode. When the effectiveness of the two types of SDs was examined separately, results showed that the active shutter-glasses SD resulted in faster responses and task completion times than the passive polarised SD, though most of the differences failed to reach statistical significance. Perceived workload when interacting with the two SD systems did not differ significantly between the active versus passive display types or between the 3D and 2D modes of operation; however, participants reported more severe discomfort after interacting with the passive polarised SD.