基于非负张量表示的扩展LDA标注模型

由于“语义鸿沟”的存在,自动图像标注是一项极具挑战性的工作。考虑到图像低层视觉特征与高层语义概念的差异,分别从图像表示与语义建模两个方面来实现自动图像标注。在图像表示方面,提出了一种正则化约束下的非负张量表示方法,用以提取符合人眼视觉直观理解的图像高阶结构特征。在语义建模方面,提出了一种三层贝叶斯模型——扩展隐Dirichlet分配。该模型利用隐变量来实现图像与标注词的关联,并通过一种基于变分推理的期望最大值方法来估计其参数。实验结果表明,ELDA模型在大规模数据库NUS-WIDE上的标注结果相较于现有方法有了显著的提高。     

图像融合的客观质量评估

随着图像融合技术在武器自动化系统中的广泛应用,图像融合的质量评估问题,即图像融合的性能评估问题,在融合方案的选择中具有重要地位。由于人工主观的图像融合评估具有很大的工程化和自动化难度,国际上研究人员都将研究目标投向了客观图像融合质量评估的研究中。从历史发展的角度系统地回顾了客观图像融合质量评估的发展历程和研究现状,对近十年来的客观图像融合质量评估方法进行了分类和分析。同时,对于国际上几种流行的图像融合质量评估方法的性能测定思路进行了介绍。对于图像融合质量评估现状的分析和讨论有助于我们对图像融合的质量评估方法进行更深入的研究。     

Using container inspection history to improve interdiction logistics for illicit nuclear materials

Strengthening the United States' ability to prevent adversaries from smuggling nuclear materials into the country is a vital and ongoing issue. The prospect of additional countries, such as Iran, obtaining the know‐how and equipment to produce these special nuclear materials in the near future underscores the need for efficient and effective inspection policies at ports and border crossings. In addition, the reduction of defense and homeland security budgets in recent years has made it increasingly important to accomplish the interdiction mission with fewer funds. Addressing these complications, in this article, we present a novel two‐port interdiction model. We propose using prior inspection data as a low‐cost way of increasing overall interdiction performance. We provide insights into two primary questions: first, how should a decision maker at a domestic port use detection data from the foreign port to improve the overall detection capability? Second, what are potential limitations to the usefulness of prior inspection data—is it possible that using prior data actually harms decision making at the domestic port? We find that a boundary curve policy (BCP) that takes into account both foreign and domestic inspection data can provide a significant improvement in detection probability. This BCP also proves to be surprisingly robust, even if adversaries are able to infiltrate shipments during transit. © 2013 Wiley Periodicals, Inc. Naval Research Logistics 60: 433‐448, 2013     

Efficient distributions of arms‐control inspection effort

A rule that constrains decision‐makers is enforced by an inspector who is supplied with a fixed level of inspection resources—inspection personnel, equipment, or time. How should the inspector distribute its inspection resources over several independent inspectees? What minimum level of resources is required to deter all violations? Optimal enforcement problems occur in many contexts; the motivating application for this study is the role of the International Atomic Energy Agency in support of the Treaty on the Non‐Proliferation of Nuclear Weapons. Using game‐theoretic models, the resource level adequate for deterrence is characterized in a two‐inspectee problem with inspections that are imperfect in the sense that violations can be missed. Detection functions, or probabilities of detecting a violation, are assumed to be increasing in inspection resources, permitting optimal allocations over inspectees to be described both in general and in special cases. When detection functions are convex, inspection effort should be concentrated on one inspectee chosen at random, but when they are concave it should be spread deterministicly over the inspectees. Our analysis provides guidance for the design of arms‐control verification operations, and implies that a priori constraints on the distribution of inspection effort can result in significant inefficiencies. © 2003 Wiley Periodicals, Inc. Naval Research Logistics, 2004.     

运动模糊图像的运动模糊方向鉴别

曝光瞬间造成图像模糊的运动通常作为直线运动近似处理,若能找出模糊图像的运动模糊方向,并将之旋转到水平轴,则二维问题可简化为一维来处理,大大简化由模糊图像估计出运动模糊点扩散函数以及图像恢复的过程,并为图像恢复的并行计算创造有利条件。由于运动模糊降低了运动方向上图像的高频成分,沿着运动方向实施高通滤波(方向微分),可保证微分图像灰度值(绝对值)之和最小。基于此,本文利用双线性插值的方法,固定并适当选取方向微分的微元大小,构造出3×3方向微分乘子,得到了高效高精度的自动鉴别运动模糊方向的新方法,并通过数值实验进行了验证。     

基于因子分解和光束法平差从航空序列图像恢复三维结构

从航空序列图像恢复目标三维结构是航空摄影测量的基本问题。针对现有算法中精度与计算量的矛盾,提出一种基于因子分解与光束法平差的新算法。作为因子分解与光束法平差的桥梁,提出基于"R T"模式共线方程的概念,详细推导了其线性化公式。仿真实验和真实图像实验表明,新算法精度高、稳定性好,可应用于航空测绘、虚拟现实、军事侦察等工程实践。     

电子稳像中运动矢量一致性的优化算法

针对电子稳像技术传统块匹配法求运动矢量的固有缺点,采取四叉树———一种变块匹配方法计算运动矢量,使求得的运动矢量有较好的一致性.与其它变块匹配方法相比,该方法克服了其它变块匹配法的不足.实验表明,使用该法求出的运动矢量有较好的一致性和准确性.     

多重网格法求解静电像管三维电场的若干问题研究

详细讨论了多重网法数值求解三维静电场的若干数学问题,诸如三维多重网格法的限制与延拓公式推导、确定静电像管三维电子光学系统中的锥面电极、圆筒电极、球面电极、圆孔膜片电极、孔电极等的空间位置及范围的方法、以及迭代计算的逻辑尺编码方法等,为像管电子光学系统设计及优化设计奠定了良好基础.     

BM弹头雷达一维像干扰初探

根据目前反导雷达使用宽带信号对目标进行一维成像,检测目标几何特征的特点,从弹道导弹(BM)弹头雷达一维像(HRRP)特征出发,参照宽带线性调频雷达成像原理,讨论了对BM弹头HRRP可采取的几种电子干扰措施,包括对BM弹头HRRP的模拟和遮盖,给出了模拟BM弹头HRRP时所需生成的各信号特征,分析了每种遮盖干扰信号的功率利用情况,最后分析了每种干扰措施的优缺点。     

《美国牧歌》的神话原型解读

本文拟以神话原型理论为理论基础来解析菲利普.罗斯的著名作品《美国牧歌》。根据弗莱对原型的定义,指出该小说呈现了人物原型、主题原型、意象原型等。这些看似偶然的原型形成了普遍的结构关联,决定了作品的内在一致性。作者通过再现特定历史时期的原型,激发读者重新思考过去、现在和未来的联系。