基于最大相关熵准则的鲁棒度量学习算法

度量亦称距离函数,是度量空间中满足特定条件的特殊函数,一般用来反映数据间存在的一些重要距离关系.而距离对于各种分类聚类问题影响很大,因此度量学习对于这类机器学习问题有重要影响.受到现实存在的各种噪声影响,已有的各种度量学习算法在处理各种分类问题时,往往出现分类准确率较低以及分类准确率波动大的问题.针对该问题,本文提出一种基于最大相关熵准则的鲁棒度量学习算法.最大相关熵准则的核心在于高斯核函数,本文将其引入到度量学习中,通过构建以高斯核函数为核心的损失函数,利用梯度下降法进行优化,反复测试调整参数,最后得到输出的度量矩阵.通过这样的方法学习到的度量矩阵将有更好的鲁棒性,在处理受噪声影响的各种分类问题时,将有效地提高分类准确率.本文将在一些常用机器学习数据集（UCI）还有人脸数据集上进行验证实验.     

基于高斯-柯西混合模型的单幅散焦图像深度恢复方法

单幅图像场景深度的获取一直是计算机视觉领域的一个难题。使用高斯分布函数或柯西分布函数近似点扩散函数模型(PSF),再根据图像边缘处散焦模糊量的大小与场景深度之间的关系估算出深度信息,是一种常用的方法。真实世界中图像模糊的缘由千变万化,高斯分布函数以及柯西分布函数并不一定是最佳的近似模型,并且传统的方法对于图像存在阴影、边缘不明显以及深度变化比较细微的区域的深度恢复结果不够准确。为了提取更为精确的深度信息,提出一种利用高斯-柯西混合模型近似PSF的方法;然后对散焦图像进行再模糊处理,得到两幅散焦程度不同的图像;再通过计算两幅散焦图像边缘处梯度的比值估算出图像边缘处的散焦模糊量,从而得到稀疏深度图;最后使用深度扩展法得到场景的全景深度图。通过大量真实图像的测试,说明新方法能够从单幅散焦图像中恢复出完整、可靠的深度信息,并且其结果优于目前常用的两种方法。     

基于Lévy分布的柔软自适应演化采样算法

已经有相关工作将演化思想引入采样算法中,并结合 Lévy 分布提出了自适应的采样算法。针对 Lévy 分布的参数设 置和“厚尾”特性的关系进行了研究,改进了基于 Lévy 分布的演化采样算法,通过设置该分布的参数 ? 值为 1.0,1.3,1.7, 2.0,分别对应四种转移概率分布,从而增加了生成的候选样本的多样性。理论分析和实验表明,改进算法在收敛速率和精度 上优于基于高斯分布,柯西分布,对称指数分布的演化采样算法和其他自适应的演化采样算法。     

Local convergence rates of simple evolutionary algorithms withCauchy mutations

The standard choice for mutating an individual of an evolutionary algorithm with continuous variables is the normal distribution; however other distributions, especially some versions of the multivariate Cauchy distribution, have recently gained increased popularity in practical applications. Here the extent to which Cauchy mutation distributions may affect the local convergence behavior of evolutionary algorithms is analyzed. The results show that the order of local convergence is identical for Gaussian and spherical Cauchy distributions, whereas nonspherical Cauchy mutations lead to slower local convergence. As a by-product of the analysis, some recommendations for the parametrization of the self-adaptive step size control mechanism can be derived     

Use of the q-Gaussian mutation in evolutionary algorithms

This paper proposes the use of the q-Gaussian mutation with self-adaptation of the shape of the mutation distribution in evolutionary algorithms. The shape of the q-Gaussian mutation distribution is controlled by a real parameter q. In the proposed method, the real parameter q of the q-Gaussian mutation is encoded in the chromosome of individuals and hence is allowed to evolve during the evolutionary process. In order to test the new mutation operator, evolution strategy and evolutionary programming algorithms with self-adapted q-Gaussian mutation generated from anisotropic and isotropic distributions are presented. The theoretical analysis of the q-Gaussian mutation is also provided. In the experimental study, the q-Gaussian mutation is compared to Gaussian and Cauchy mutations in the optimization of a set of test functions. Experimental results show the efficiency of the proposed method of self-adapting the mutation distribution in evolutionary algorithms.     

Audio steganalysis of spread spectrum information hiding based on statistical moment and distance metric

Audio information hiding has attracted more attentions recently. Spread spectrum (SS) technique has developed rapidly in this area due to the advantages of good robustness and immunity to noise attack. Accordingly detecting the SS hiding effectively and verifying the presence of the secrete message are important issues. In this paper we present two steganalysis algorithms for SS hiding. Both the two methods are based on machine learning theory and discrete wavelet transform (DWT). In the algorithm I, we introduce Gaussian mixture model (GMM) and generalize Gaussian distribution (GGD) to character the probability distribution of wavelet sub-band. Then the absolute probability distribution function (PDF) moment is extracted as feature vectors. In the algorithm II, we propose distance metric between GMM and GGD of wavelet sub-band to distinguish cover and stego audio. Four distance metrics (Kullback-Leibler Distance, Bhattacharyya Distance, Earth Mover’s Distance, L2 Distance) are calculated as feature vectors. The support vector machine (SVM) classifier is utilized for classification. The experiment results of both two proposed algorithms can achieve better detecting performance. Even when embedding strength gets 0.0005, the correct detection rate can reach up to 90%. Its simplicity and extensibility indicate further application in other audio steganalysis.     

Noise analysis and synthesis for 3D laser depth scanners

This paper analyses the noise present in range data measured by a Konica Minolta Vivid 910 scanner, in order to better characterise real scanner noise. Methods for denoising 3D mesh data have often assumed the noise to be Gaussian, and independently distributed at each mesh point. We show via measurements of an accurately machined almost planar test surface that real scanner data does not have such properties: the errors are not quite Gaussian, and more importantly, exhibit significant short range correlation. We use this to give a simple model for generating noise with similar characteristics. We also consider how noise varies with such factors as laser intensity, orientation of the surface, and distance from the scanner. Finally, we evaluate the performance of three typical mesh denoising algorithms using real and synthetic test data, and suggest that new denoising algorithms are required for effective removal of real noise.     

Registering multiple range images based on distance metric of surface-to-surface

1 Introduction In recent years, there has been growing interest in the range sensing techniques for building 3D computer models of real-world objects and scenes without requiring hu-mans to manually produce these models using laborious and error-prone CAD-based approaches. Using range sensors, users are able to capture 3D range images of objects from different viewpoints that may be combined to form the final model of the object or scene[1]. These models then may be used for a variety of app…     

Application of feature space trajectory classifier to identification of multi-aspect radar signals

In this paper, a feature space trajectory (FST) classifier is applied to identify an unknown radar target. To improve the identification accuracy, we make use of information at multiple aspects of a radar target, and the FST classifier is combined with two different rules: majority vote and sum vote. In addition, two different algorithms via the simultaneous use of FST concept and line-to-line distance metric are presented to classify multi-aspect radar signals. Experimental results show that the proposed two algorithms significantly outperform the traditional FST classifier combined with majority vote and sum vote.     

Adjustable SAD matching algorithm using frequency domain

Fast Fourier transforms (FFTs) which are O(N logN) algorithms to compute a discrete Fourier transform (DFT) of size N have been called one of the ten most important algorithms of the twentieth century. However, even though many algorithms have been developed to speed up the computation the sum of absolute difference (SAD) matching, they are exclusively designed in the spatial domain. In this paper, we propose a fast frequency algorithm to speed up the process of (SAD) matching. We use a new approach to approximate the SAD metric by cosine series which can be expressed in correlation terms. These latter can be computed using FFT algorithms. Experimental results demonstrate the effectiveness of our method when using only the first correlation terms for block and template matching in terms of accuracy and speed. The proposed algorithm is suitable for software implementations and has a deterministic execution time unlike the existing fast algorithms for SAD matching.

基于FPGA的双目立体视觉系统

立体视觉的目的之一就是为了获得周围场景的3维信息,其关键在于匹配算法。然而即便是使用目前先进的通用处理器,其计算致密视差图所需的时间仍无法满足高速自主导航的需求。为了解决这个问题,提出了一种基于现场可编程门阵列(FPGA)的双目立体视觉系统的设计方案,同时介绍了系统的硬件结构,并在讨论区域匹配的快速算法的基础上,提出了基于FPGA的像素序列和并行窗口算法框架,用以实现零均值像素灰度差平方和(ZSSD)的匹配算法。该算法是先将视频信号经解码芯片生成场景立体图像对,并由FPGA来完成立体图像对的几何校正和ZSSD匹配算法,然后将获得的致密视差图通过PC I总线发送至上位机。实践表明,该算法效果好、速度快,不仅具有较强的鲁棒性,并且硬件系统性能稳定、可靠。此外,该方案还适用于像素灰度差的绝对值和(SAD)和像素灰度差的平方和(SSD)等多种传统区域匹配算法的快速实现和实时处理。     

结合缺失模式的不完整数据模糊聚类

数据的完整性是数据可用性的重要维度。由于数据采集等过程中存在的问题,现实中的数据往往存在缺失。现有的聚类算法在面对不完整数据时一般采用忽略缺失或填补缺失的策略,但是当数据缺失属于非随机缺失时,这样的处理策略会导致聚类精度严重下降。当数据缺失属于非随机缺失时,数据缺失模式与缺失属性的取值相关,因此在不完整对象的相似度量中加入缺失模式相似的度量,提出了两种结合缺失模式的 PCM(Possibilistic c-means)模糊聚类算法:最小化缺失模式距离之和的 PatDistPCM 算法和基于缺失模式聚类的 PatCluPCM 算法。在两个公开数据集上的实验证明, 考虑缺失模式的模糊聚类PatDistPCM和PatCluPCM算法,在对存在非随机缺失的数据进行聚类时 ,能有效提高聚类结果的准确性。     

Gaussian coordinate systems for the Kerr metric

We present a class of Gaussian coordinate systems for the Kerr metric obtained from the relativistic Hamilton-Jacobi equation. We discuss the Cauchy problem of such a coordinate system. In the appendix, we present the JEK (Jordan-Ehlers-Kundt) formulation of General Relativity—the so-called quasi-Maxwell equations—which acquires a simpler form in the Gaussian coordinate system. We show how this set of equations can be used to regain the internal metric of the Schwarzschild solution and, with this in mind, we suggest a possible way to find out a physically significant internal solution for the Kerr metric.     

Multidimensional mutations in evolutionary algorithms based on real-valued representation

This work is focused on the fact that the most probable distance of mutated points in multi-dimensional Gaussian and Cauchy mutations is not in a close neighborhood of the origin, but at a certain distance from it. In the case of the Gaussian mutation, this distance is proportional to the norm of the standard deviation vector and increases with the landscape dimension. This may cause a decrease in the sensitivity of the evolutionary algorithm to narrow peaks when the landscape dimension increases, but, simultaneously, it strengthens the exploration property of the algorithm. Moreover, the influence of the reference frame orientation on the effectiveness of the non-spherical multi-dimensional Cauchy mutation is analyzed using simulation experiments. Four multi-dimensional mutations (Gaussian, modified Gaussian, non-spherical and spherical Cauchy mutations) are applied to two classes of evolutionary algorithms based on real-valued representation, i.e. Galar's evolutionary search with soft selection and evolutionary programming. A comparative analysis is provided for convergence to the local optimum, sensitivity to narrow peaks, saddle crossing and symmetry problems.     

Pruning using parameter and neuronal metrics

In this article, we introduce a measure of optimality for architecture selection algorithms for neural networks: the distance from the original network to the new network in a metric defined by the probability distributions of all possible networks. We derive two pruning algorithms, one based on a metric in parameter space and the other based on a metric in neuron space, which are closely related to well-known architecture selection algorithms, such as GOBS. Our framework extends the theoretically range of validity of GOBS and therefore can explain results observed in previous experiments. In addition, we give some computational improvements for these algorithms.     

融合距离度量和高斯混合模型的中文词义归纳模型

词义归纳是解决词义知识获取的重要研究课题,利用聚类算法对词义进行归纳分析是目前最广泛采用的方法。通过比较K-Means聚类算法和EM聚类算法在 各自 词义归纳模型上的优势,提出一种新的融合距离度量和高斯混合模型的聚类算法,以期利用两种聚类算法分别在距离度量和数据分布计算上的优势,挖掘数据的几何特性和正态分布信息在词义聚类分析中的作用,从而提高词义归纳模型的性能。实验结果表明,所提混合聚类算法对于改进词义归纳模型的性能是十分有效的。     

非贪婪的鲁棒性度量学习算法

目的 度量学习是机器学习与图像处理中依赖于任务的基础研究问题。由于实际应用背景复杂,在大量不可避免的噪声环境下,度量学习方法的性能受到一定影响。为了降低噪声影响,现有方法常用L1距离取代L2距离,这种方式可以同时减小相似样本和不相似样本的损失尺度,却忽略了噪声对类内和类间样本的不同影响。为此,本文提出了一种非贪婪的鲁棒性度量学习算法——基于L2/L1损失的边缘费歇尔分析（marginal Fisher analysis based on L2/L1 loss,MFA-L2/L1）,采用更具判别性的损失,可提升噪声环境下的识别性能。方法 在边缘费歇尔分析（marginal Fisher analysis,MFA）方法的基础上,所提模型采用L2距离刻画相似样本损失、L1距离刻画不相似样本损失,同时加大对两类样本的惩罚程度以提升方法的判别性。首先,针对模型非凸带来的求解困难,将目标函数转为迭代两个凸函数之差便于求解;然后,受DCA（difference of convex functions algorithm）思想启发,推导出非贪婪的迭代求解算法,求得最终度量矩阵;最后,算法的理论证明保证了迭代算法的收敛性。结果 在5个UCI（University of California Irrine）数据集和7个人脸数据集上进行对比实验：1）在不同程度噪声的5个UCI数据集上,MFA-L2/L1算法最优,且具有较好的抗噪性,尤其在30%噪声程度的Seeds和Wine数据集上,与次优方法LDA-NgL1（non-greedy L1-norm linear discriminant analysis））相比,MFA-L2/L1的准确率高出9%;2）在不同维度的AR和FEI人脸数据集上的实验,验证了模型采用L1损失、采用L2损失提升了模型的判别性;3）在Senthil、Yale、ORL、Caltech和UMIST人脸数据集的仿真实验中,MFA-L2/L1算法呈现出较强鲁棒性,性能排名第1。结论 本文提出了一种基于L2/L1损失的鲁棒性度量学习模型,并推导了一种便捷有效的非贪婪式求解算法,进行了算法收敛性的理论分析。在不同数据集的不同噪声情况下的实验结果表明,所提算法具有较好的识别率和鲁棒性。     

闪烁噪声下基于模型集合切换的机动目标跟踪算法

在目标跟踪过程中,由于目标与传感器相对位置的变化以及目标不同部位的反射干扰,传感器测量数据中的测量噪声不再是严格意义上的高斯噪声,而变为具有长拖尾特性的闪烁噪声,而闪烁噪声的出现将严重影响线性卡尔曼滤波器的滤波性能。机动目标跟踪算法应同时考虑目标运动模式变化和闪烁噪声出现对滤波过程的影响。将滤波方程集合划分为包含不同目标运动模式的模型集合,提出了基于模型集合切换的机动目标跟踪(MSSM)算法。仿真结果验证了MSSM算法的有效性。     

Robust kernelized approach to clustering by incorporating new distance measure

A new data clustering algorithm Density oriented Kernelized version of Fuzzy c-means with new distance metric (DKFCM-new) is proposed. It creates noiseless clusters by identifying and assigning noise points into separate cluster. In an earlier work, Density Based Fuzzy C-Means (DOFCM) algorithm with Euclidean distance metric was proposed which only considered the distance between cluster centroid and data points. In this paper, we tried to improve the performance of DOFCM by incorporating a new distance measure that has also considered the distance variation within a cluster to regularize the distance between a data point and the cluster centroid. This paper presents the kernel version of the method. Experiments are done using two-dimensional synthetic data-sets, standard data-sets referred from previous papers like DUNN data-set, Bensaid data-set and real life high dimensional data-sets like Wisconsin Breast cancer data, Iris data. Proposed method is compared with other kernel methods, various noise resistant methods like PCM, PFCM, CFCM, NC and credal partition based clustering methods like ECM, RECM, CECM. Results shown that proposed algorithm significantly outperforms its earlier version and other competitive algorithms.     

Block-matching algorithm based on differential evolution for motion estimation

Motion estimation is one of the major problems in developing video coding applications. Among all motion estimation approaches, block-matching (BM) algorithms are the most popular methods due to their effectiveness and simplicity for both software and hardware implementations. A BM approach assumes that the movement of pixels within a defined region of the current frame (macro block, MB) can be modeled as a translation of pixels contained in the previous frame. In this procedure, the motion vector is obtained by minimizing the sum of absolute differences (SAD) produced by the MB of the current frame over a determined search window from the previous frame. The SAD evaluation is computationally expensive and represents the most consuming operation in the BM process. The most straightforward BM method is the full search algorithm (FSA), which finds the most accurate motion vector, exhaustively calculating the SAD values for all the elements of the search window. Over this decade, several fast BM algorithms have been proposed to reduce the number of SAD operations by calculating only a fixed subset of search locations at the cost of poor accuracy. In this paper, a new algorithm based on differential evolution (DE) is proposed to reduce the number of search locations in the BM process. To avoid computing several search locations, the algorithm estimates the SAD values (fitness) for some locations using the SAD values of previously calculated neighboring positions. As the proposed algorithm does not consider any fixed search pattern or any other different assumption, a high probability for finding the true minimum (accurate motion vector) is expected. In comparison with other fast BM algorithms, the proposed method deploys more accurate motion vectors, yet delivering competitive time rates.