基于Mean Shift的相似性变换和仿射变换目标跟踪算法

传统的Mean Shift(MS)算法只能对发生平移和尺度变化的目标进行跟踪,而对于具有相似性变换或者更复杂的仿射变换的目标跟踪效果很不理想或无法跟踪。为了解决这一问题,提出了两种基于MS的改进算法。第一种算法针对仿射变换,根据奇异值分解理论,仿射变换矩阵可以分解成两个旋转矩阵和一个对角矩阵的乘积,在此基础上建模了一种新的候选目标模型。通过Bhattacharyya系数将目标跟踪问题转化成以仿射变换参数为变量的最优化问题,推导相关参量的一阶偏导数并令其为零从而得出相对于仿射变换的MS算法。另外,针对进行相似性变换的目标也提出了一种新的候选目标模型,并用类似的梯度下降算法估计目标的平移向量和旋转角度。实验结果表明,提出的算法能够跟踪具有相似性变换或仿射变换的目标,比传统的MS算法具有更好的跟踪性能。

结合双向相似性变换的重定向图像质量评价

目的显示设备的多样化使得图像重定向的作用日益凸显。不同的重定向方法产生不同视觉感受的重定向图像,而如何评价重定向图像的质量,优化重定向算法是当前研究的热点与难点,为此,提出一种结合双向相似性变换的重定向图像质量评价方法。方法首先对原始图像和重定向图像进行像素点双向匹配,利用网格顶点坐标对计算前向变换矩阵和后向变换矩阵。然后由相似性变换矩阵与标准变换矩阵间的距离得到重定向图像的几何失真。由网格面积缺失得到重定向图像的信息损失。最后结合网格的显著性,融合前向匹配与后向匹配的几何失真和信息损失得到重定向图像的质量。结果该方法在Retarget Me和CUHK数据库上的KRCC(Kendall rank correlation coefficient)和SROCC(Spearman rank-order correlation coefficient)性能分别达到了0.46和0.71,较现有方法有较大提升。在前向匹配与后向匹配测试中,双向匹配的测试结果优于单向匹配。结论本文方法将图像的重定向处理看做相似性变换过程。实验结果表明,从相似性变换矩阵中提取的相关特征能够较精确度量重定向图像的几何失真,而由此引发的网格面积缺失也能准确反映出重定向图像的信息损失。另外,采用双向匹配机制一定程度上减少了像素匹配误差对实验结果的影响,有效提升了重定向图像质量预测的准确性。该方法对重定向图像的质量评价效果好,适用于重定向图像的质量预测及算法优化。

基于线性动态系统稀疏编码的异常事件检测

线性动态系统模型结合稀疏编码实现异常事件检测。线性动态系统可有效地捕捉动态纹理在时间和空间的转移信息,描述视频的时空小块。然而,线性动态系统属于非欧氏空间,无法直接用传统的稀疏编码进行异常检测。基于约束凸优化公式,将相似性变换与稀疏编码结合,可实现线性动态系统稀疏编码的优化求解。实验表明,所提出的方法具有更好的性能。

基于颜色不变量和SURF的彩色图像拼接

提出了一种基于颜色不变量和SURF算法相结合的彩色图像拼接方法。该方法利用图像彩色信息计算得到的颜色不变量信息代替灰度信息作为输入,提取图像SURF特征点并进行特征点匹配,根据相似性变换原理对误匹配点进行过滤,提高变换矩阵计算的准确率,采用亮度渐变原则对重叠区域进行像素平滑过渡,实现图像无缝拼接。实验结果表明,该方法在保持算法的快速性和准确性的同时,获得的配准点多而且准确,采用亮度渐变原则拼接增强了对光照变化的鲁棒性,且有效地消除了拼接痕迹,在图像拼接和地质分析领域有一定的实用价值。

一种基于浮点数编码遗传算法的颅像叠加方法

针对计算机辅助颅像重合技术中三维颅骨模型与二维人脸照片的颅像叠加问题,提出一种基于浮点数编码遗传算法的自动叠加方法。首先使用遗传算法寻找三维颅骨相对于二维照片上对应标志点的最佳匹配几何变换f,遗传算法采用浮点数编码,选择、交叉、变异算子分别使用最佳保留选择算子、启发式交叉算子和随机变异算子,然后通过相似性变换和透视投影变换,将三维颅骨模型映射到二维人脸照片上,实现最佳颅像叠加。实验表明,此方法实现了自动化的颅像叠加,且耗时短,准确度高。

Classification of forearm action surface EMG signals based on fractal dimension

Surface electromyogram （EMG） signals were identified by fractal dimension.Two patterns of surface EMG signals were acquired from 30 healthy volunteers＇ right forearm flexor respectively in the process of forearm supination （FS） and forearm pronation （FP）.After the raw action surface EMG （ASEMG） signal was decomposed into several sub-signals with wavelet packet transform （WPT）,five fractal dimensions were respectively calculated from the raw signal and four sub-signals by the method based on fuzzy self-similarity.The results show that calculated from the sub-signal in the band 0 to 125 Hz,the fractal dimensions of FS ASEMG signals and FP ASEMG signals distributed in two different regions,and its error rate based on Bayes decision was no more than 2.26%.Therefore,the fractal dimension is an appropriate feature by which an FS ASEMG signal is distinguished from an FP ASEMG signal.

Fast Digital Correlation Calculation Method in Extended Beacon Wavefront Detection

On the basis of Hartmann Shack sensor imaging analysis, a new method is presented with which the wavefront slope can be obtained when the object is incoherent and extended. This method, which is demonstrated by both theoretical interpreting and computer simulation, explains how to measure the wavefront slope difference between two sub apertures through the determination of image displacements on detector plane. It includes a fast and accurate digital algorithm for detecting wavefront disturbance, which is much suitable for realization in such electrical hardwares as digital signal processors.

基于稀疏编码的动态纹理识别

目的线性动态系统有效地捕捉了动态纹理在时间和空间的转移信息。然而,线性动态系统属于非欧氏空间模型,无法直接应用传统的稀疏编码进行分类识别,为此提出一种基于稀疏编码线性动态系统的求解方法并应用于动态纹理识别。方法基于约束凸优化公式,将稀疏编码和控制论中相似性变换结合,优化学习模型参数,解决应用稀疏编码进行分类识别的问题,实现有效的动态纹理识别。结果在公开的动态纹理图像数据库UCLA上进行实验并与其他方法进行比较,实验结果表明,本文方法具有更好的性能,识别率可达97%,且对遮挡具有更好的鲁棒性。结论本文方法对动态纹理及遮挡情况具有更好的识别率。

Classification of hyperspectral remote sensing images using frequency spectrum similarity

An algorithm of hyperspectral remote sensing images classification is proposed based on the frequency spectrum of spectral signature.The spectral signature of each pixel in the hyperspectral image is taken as a discrete signal,and the frequency spectrum is obtained using discrete Fourier transform.The discrepancy of frequency spectrum between ground objects' spectral signatures is visible,thus the difference between frequency spectra of reference and target spectral signature is used to measure the spectral similarity.Canberra distance is introduced to increase the contribution from higher frequency components.Then,the number of harmonics involved in the proposed algorithm is determined after analyzing the frequency spectrum energy cumulative distribution function of ground object.In order to evaluate the performance of the proposed algorithm,two hyperspectral remote sensing images are adopted as experimental data.The proposed algorithm is compared with spectral angle mapper (SAM),spectral information divergence (SID) and Euclidean distance (ED) using the product accuracy,user accuracy,overall accuracy,average accuracy and Kappa coefficient.The results show that the proposed algorithm can be applied to hyperspectral image classification effectively.

Spatiotemporal Rogue Waves for the Variable-Coefficient (3+1)-Dimensional Nonlinear Schrdinger Equation

With the help of the similarity transformation connected the variable-eoeicient （3＋1）-dimensionai nonlin- ear Sehroedinger equation with the standard nonlinear Schr6dinger equation, we firstly obtain first-order and second-order rogue wave solutions. Then, we investigate the controllable behaviors of these rogue waves in the hyperbolic dispersion decreasing profile. Our results indicate that the integral relation between the accumulated time T and the reai time t is the basis to realize the control and manipulation of propagation behaviors of rogue waves, such as sustainment and restraint. We can modulate the value To to achieve the sustained and restrained spatiotemporai rogue waves. Moreover, the controllability for position of sustainment and restraint for spatiotemporai rogue waves can aiso be realized by setting different values of Xo.

Exact Solutions of the Two-Dimensional Cubic-Quintic Nonlinear Schrdinger Equation with Spatially Modulated Nonlinearities

Applying the similarity transformation, we construct the exact vortex solutions for topological charge S≥1 and the approximate fundamental soliton solutions for S = 0 of the two-dimensional cubic-quintic nonlinear Schrodinger equation with spatially modulated nonlinearities and harmonic potential. The linear stability analysis and numerical simulation are used to exam the stability of these solutions. In different profiles of cubic-quintic nonlinearities, some stable solutions for S 〉 0 and the lowest radial quantum number n = 1 are found. However, the solutions for n ≥ 2 are all unstable.

A property of Pisot numbers and Fourier transforms of self-similar measures

For any Pisot number β it is known that the set F（β） ={t ： limn→∞‖tβn‖ = 0} is countable, where ‖α‖ is the distance between a real number a and the set of integers. In this paper it is proved that every member in this set is of the form cβn, where n is a nonnegative integer and e is determined by a linear system of equations. Furthermore, for some self-similar measures μ associated with β, the limit at infinity of the Fourier transforms limn→μ（tβn）≠0 if and only if t is in a certain subset of F（β）. This generalizes a similar result of Huang and Strichartz.