基于结构化遮挡编码和极限学习机的局部遮挡人脸识别

提出使用结构化遮挡编码（SOC）结合极限学习机（ELM）的算法来处理人脸识别中的遮挡问题。首先，使用SOC去除图像上的遮挡物，将遮挡物体与人脸分离开；同时，通过局部性约束字典（LCD）来估计遮挡物的位置，建立遮挡字典和人脸字典。然后，将建立好的人脸字典矩阵进行归一化处理，并利用ELM对归一化的数据进行分类识别。最后，在AR人脸库上进行的仿真实验结果表明，所提方法对不同遮挡物和不同区域遮挡的图像具有较好的识别率和鲁棒性。     

NLRP3 Inflammasome Activation Is Involved in LPA1-Mediated Brain Injury after Transient Focal Cerebral Ischemia

Lysophosphatidic acid receptor 1 (LPA₁) contributes to brain injury following transient focal cerebral ischemia. However, the mechanism remains unclear. Here, we investigated whether nucleotide-binding oligomerization domain-like receptor family pyrin domain containing 3 (NLRP3) inflammasome activation might be an underlying mechanism involved in the pathogenesis of brain injury associated with LPA₁ following ischemic challenge with transient middle cerebral artery occlusion (tMCAO). Suppressing LPA₁ activity by its antagonist attenuated NLRP3 upregulation in the penumbra and ischemic core regions, particularly in ionized calcium-binding adapter molecule 1 (Iba1)-expressing cells like macrophages of mouse after tMCAO challenge. It also suppressed NLRP3 inflammasome activation, such as caspase-1 activation, interleukin 1β (IL-1β) maturation, and apoptosis-associated speck-like protein containing a caspase recruitment domain (ASC) speck formation, in a post-ischemic brain. The role of LPA₁ in NLRP3 inflammasome activation was confirmed in vitro using lipopolysaccharide-primed bone marrow-derived macrophages, followed by LPA exposure. Suppressing LPA₁ activity by either pharmacological antagonism or genetic knockdown attenuated NLRP3 upregulation, caspase-1 activation, IL-1β maturation, and IL-1β secretion in these cells. Furthermore, nuclear factor-κB (NF-κB), extracellular signal-regulated kinase 1/2 (ERK1/2), and p38 were found to be LPA₁-dependent effector pathways in these cells. Collectively, results of the current study first demonstrate that LPA₁ could contribute to ischemic brain injury by activating NLRP3 inflammasome with underlying effector mechanisms.     

超音速火焰喷涂镍基涂层颗粒沉积特性的数值模拟

目的 考虑后续不同粒径颗粒随机冲击的影响,探索热喷涂涂层颗粒的沉积特性。方法 利用ABAQUS建立颗粒与基底冲击模型,通过颗粒冲击的凹坑深度和应力分布进行网格收敛性研究。通过实验验证模型的可行性。随后,应用验证模型研究颗粒以不同入射角和速度冲击基底时的沉积特性,以及4个颗粒重叠冲击基底及多颗粒随机冲击基底表面时的沉积特性。结果 在颗粒入射角从15°增至60°时,颗粒更好地附着于基底表面;当颗粒速度从350 m/s增至500 m/s时发生了溅射现象,可能造成绝热剪切失稳现象,形成有效结合;在4个颗粒冲击基底时,第2个颗粒对第1个颗粒及基底的影响都最明显;当多颗粒随机冲击基底时,在后续颗粒的冲击和沉积作用下,填充颗粒的形状不规则,同时第1层颗粒可能与基底形成机械咬合。结论 在超音速火焰喷涂时应当倾斜一定角度,同时提升颗粒速度,这对制备涂层更有利;在颗粒重叠冲击时,后续颗粒增大了第1个颗粒的压缩效果,且更深入地嵌入不锈钢基底,这有利于颗粒与颗粒之间的后续黏结;当多颗粒随机冲击基底时,在第1层沉积颗粒与基底之间,以及涂层内相邻颗粒之间均观察到高塑性应变,表明涂层出现黏结,同时后期沉积的颗粒未完整压缩变形。     

多尺度相似性迭代查找的可靠双目视差估计

目的 双目视差估计可以实现稠密的深度估计,因而具有重要研究价值。而视差估计和光流估计两个任务之间具有相似性,在两种任务之间可以互相借鉴并启迪新算法。受光流估计高效算法RAFT （recurrent all-pairs field transforms）的启发,本文提出采用单、双边多尺度相似性迭代查找的方法实现高精度的双目视差估计。针对方法在不同区域估计精度和置信度不一致的问题,提出了左右图像视差估计一致性检测提取可靠估计区域的方法。方法 采用金字塔池化模块、跳层连接和残差结构的特征网络提取具有强表征能力的表示向量,采用向量内积表示像素间的相似性,通过平均池化得到多尺度的相似量,第0次迭代集成初始视差量,根据初始视差单方向向左查找多尺度的相似性得到的大视野相似量和上下文3种信息,而其他次迭代集成更新的视差估计量,根据估计视差双向查找多尺度的相似性得到的大视野相似量和上下文3种信息,集成信息通过第0次更新的卷积循环神经网络和其他次更新共享的卷积循环神经网络迭代输出视差的更新量,多次迭代得到最终的视差估计值。之后,通过对输入左、右图像反序和左右翻转估计右图视差,对比左、右图匹配点视差差值的绝对值和给定阈值之差判断视差估计置信度,从而实现可靠区域提取。结果 本文方法在Sceneflow数据集上得到了与先进方法相当的精度,平均误差只有0.84像素,并且推理时间有相对优势,可以和精度之间通过控制迭代次数灵活平衡。可靠区域提取后,Sceneflow数据集上误差进一步减小到了历史最佳值0.21像素,在KITTI （Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago）双目测试数据集上,估计区域评估指标最优。结论 本文方法对于双目视差估计具有优越性能,可靠区域提取方法能高效提取高精度估计区域,极大地提升了估计区域的可靠性。     

基于椭圆拟合的车辆遮挡处理算法

为了对遮挡车辆进行分割,提出了一种基于椭圆拟合的车辆遮挡处理算法。在遮挡检测方面,利用最小二乘椭圆拟合法检测独立前景区域中的车尾阴影。如果检测到2个车尾阴影,则判断前景区域中存在着遮挡现象。在遮挡处理方面,利用与车道线方向平行的直线将前景区域轮廓分割为2个子轮廓。之后,提出了长轴方向固定的最小二乘椭圆拟合法对2个子轮廓分别进行拟合,并定义2个子轮廓到各自椭圆的距离之和作为分割评价函数,最后选择具有最小评价值的直线作为车辆遮挡的分割直线。实验结果表明,该算法能够准确地检测和分割车辆间的遮挡现象。     

Identification and Functional Analysis of MicroRNAs in Mice following Focal Cerebral Ischemia Injury

Numerous studies have demonstrated that genes, RNAs, and proteins are involved in the occurrence and development of stroke. In addition, previous studies concluded that microRNAs (miRNAs or miRs) are closely related to the pathological process of ischemic and hypoxic disease. Therefore, the aims of this study were to quantify the altered expression levels of miRNAs in the infarct region 6 h after middle cerebral artery occlusion (MCAO)-induced focal cerebral ischemia in mice using a large-scale miRNAs microarray. Firstly, MCAO-induced cerebral ischemic injuries were investigated by observing the changes of neurological deficits, infarct volume and edema ratio. One hundred and eighteen differentially expressed miRNAs were identified in the infarct region of mice following the MCAOs compared with sham group (p < 0.05 was considered as significant). Among these 118 significantly expressed microRNAs, we found that 12 miRNAs were up-regulated with fold changes lager than two, and 18 miRNAs were down-regulated with fold changes less than 0.5 in the infarct region of mice following the 6 h MCAOs, compared with the sham group. Then, these 30 miRNAs with expression in fold change larger than two or less than 0.5 was predicted, and the functions of the target genes of 30 miRNAs were analyzed using a bioinformatics method. Finally, the miRNA-gene network was established and the functional miRNA-mRNA pairs were identified, which provided insight into the roles of the specific miRNAs that regulated specified genes in the ischemic injuries. The miRNAs identified in this study may represent effective therapeutic targets for stroke, and further study of the role of these targets may increase our understanding of the mechanisms underlying ischemic injuries.     

基于块的Mean-shift跟踪算法

针对传统Mean-shift跟踪算法在目标发生遮挡和形态变化时跟踪性能下降的缺点,提出了一种基于块的Mean-shift跟踪算法,该算法主要特点有：(1)将跟踪目标平均分块,每小块独立进行传统Mean-shift跟踪,利用小块跟踪未被遮挡的目标部分;(2)跟踪检测器检测目标小块跟踪的有效性,筛选出无效跟踪的目标小块,解决了目标分块造成跟踪性能下降的问题;(3)归一化互相关检测器和邻域一致检测增加了对目标空间信息的检测,弥补了Mean-shift算法的局限性,增加了跟踪的鲁棒性。实验表明,该算法在目标发生遮挡和形态变化时仍然可以有效的实现跟踪。     

玻璃体腔注射曲安奈德治疗视网膜分支静脉阻塞继发黄斑囊样水肿疗效观察

观察玻璃体腔注射曲安奈德(intravitreal triamcinolone acetonide,IVTA)治疗视网膜分支静脉阻塞(brach retinal vein occlusion,BRVO)继发黄斑囊样水肿(cystoid macular edema,CME)的临床效果.对33例(34眼)BRVO继发CME...     

正投影模型下基于一维子空间的遮挡点恢复方法

为了恢复图像遮挡点,假设相机为正投影模型,提出了一种基于1维子空间的遮挡点恢复方法,该方法利用第1幅图像点构成的2个行向量外加1个行向量可以组成三维空间点构生成的子空间的特性,线性迭代地求取遮挡点的真实图像位置.此方法最大的优点是在整个求解过程中都是线性的.收敛性实验表明了该方法具有很好的收敛性能,同时,和Eriksson方法及Ma方法比较显示了本文方法具有较好的收敛精度.     

Contour‐based polygonal ambient occlusion using a single‐depth texture

We present a novel approximation of a global illumination technique called contour‐based polygonal ambient occlusion, an approach that darkens by only using a depth texture as screen‐space ambient occlusion without additional information such as a normal buffer. We introduce a discrete level structure to calculate the amount of occlusion by using the contour level, an integer value derived from the depth difference of neighboring texels. We use the uniform sampling to define the positions of the neighboring texels in the depth texture. Our method does not depend on geometric complexity because our method works in screen space and works well for both static and dynamic scenes without any precomputation. Copyright © 2013 John Wiley & Sons, Ltd.