提出一种针对非线性动力方程的改进精细积分方法。该方法是在时间步长内采用分段的三次样条函数拟合非齐次项,保持高精度拟合的同时避免了求导运算和高次多项式插值带来的Runge现象。通过引入4×2个变量将动力方程增加四维转化为齐次方程,并建立相应的通解格式,避免了状态空间下系统矩阵求逆。将指数矩阵分为四个子模块,利用各模块的特点分别进行理论推导及基于精细积分法进行分步、分块计算得到相应的理论解和高精度数值解,无需反复计算整个指数矩阵,提高了解算效率。针对含未知状态量的非齐次项,引入预测-校正的方法进行迭代求解。数值计算结果表明了本文方法的有效性。

An improved precise integration method for nonlinear dynamic equations was proposed in this paper.The inhomogeneous terms were fitted by employing piecewise cubic spline interpolation at ever time-step to increase fitting precision and to avoid calculating differential coefficient,curve oscillation of high-order polynomial fitting was also avoided.The original nonlinear dynamic equations were converted into homogenous equations with four increased dimensions by importing 4×2 variables,the relevant cal-culating format was built without inversing the state-space system matrix.Based on the mathematic fea-ture of homogenized matrix,a partitioning and step by step precise integration calculating method was presented by theoretical deducing and blocked calculation,no need for computing matrix exponential at every time step,and the efficiency was improved.Predict-correct method based on Taylor format was used for inhomogeneous terms with state variables.The numerical examples show

针对机器视觉检测TFT-LCD Mura缺陷时存在的图像整体亮度不均匀、背景复杂等影响检测准确性的问题,提出一种基于B样条曲面拟合的背景抑制方法。在最小二乘法准则的约束下,采用双三次B样条曲面拟合算法拟合出背景,并添加光顺项调整拟合精度,用原始图像减去拟合背景,从而消除亮度不均匀背景对缺陷分割造成的影响。为提高算法速度,对原始图像进行分块拟合,并将双三次B样条函数分解为一元函数求解,减小了计算量,同时避免了对原函数求解时容易出现的病态解问题。实验结果表明,该算法准确、高效。

A background suppression method based on B-spline surface fitting was proposed to solve the problem of the accuracy of Mura detection, which is often caused by the uneven brightness and complex background when employing machine vision on Mura detection of TFT-LCD. Under the constraint of least square criterion, bicubic B-spline surface fitting algorithm was deployed to fit the uneven brightness background and a fairing item was added to adjust fitting precision. In order to increase the performance of the proposed algorithm and decrease the calculation time, divided fitting method was used on original image and the bicubic B-spline function was analyzed into one-dimensional functions for efficiently solving. Furthermore, pathological problems caused by solving of the two-dimensional functions could be avoided simultaneously. The effect on the Mura defect segmentation caused by uneven brightness background was eliminated by subtracting the fitted background from the original image. The

感知哈希（Perceptual Hashing）是多媒体数据集到摘要集的单向映射，为多媒体数字内容的标识、检索、认证等应用提供了安全可靠的技术支撑．本文提出一种融合视觉感知及时空域特征的视频感知哈希算法．算法首先对视频序列每一帧进行随机可重叠分块，并计算每个分块以像素为单位的亮度均值，在某一步长下，以同一帧的分块亮度差作为视频帧空域特征，以不同视频帧相同位置的分块亮度差作为时域特征，通过哈希量化得到时空域感知哈希，通过时空域感知哈希融合，最后得到简洁的视频唯一标识——摘要哈希．实验结果表明，该算法表现出较好的鲁棒性与区分性，通过相似度拟合图算法分析，可以实现视频篡改的准确检测及定位．

Perceptual hashing is a one-directional mapping from multimedia data sets to the abstract sets ,which provides a safe and reliable tool for identification ,retrieval ,and authentication of multimedia digital contents .This paper proposes a video per-ceptual hashing algorithm fuseing visual perception and spatiotemporal features .The algorithm first randomly divides each video frame into a number of overlapping blocks ,and then calculates the average brightness (AB) of each block .While spatial features are described by differences of the AB among blocks in the same frame ,the features in time domain is represented by the differences of the same bock across time .Finally ,the two features are quantized and fused to generate the unique id ,i .e .abstract hashing for each video .The experimental results show that the features are robust ,discriminative ,and can effectively detect and accurately locate video tampering using correlation analysis .