基于BoC-BoF特征的图像检索方法研究

为了优化基于内容的图像检索方法,提出了一种融合特征来表征图像内容.首先,提取基于RootSift描述子的特征词袋(Bag-of-Features,BoF)表示向量,获得图像的边缘和形状信息;其次,采用基于HSV的颜色词袋(Bag-of-Colors,BoC)表示向量来代替传统颜色直方图方法,获取图像的颜色信息;最后,将BoF表示向量和BoC表示向量相融合,形成BoC-BoF特征向量.BoC-BoF特征有效地实现了全局特征和局部特征的融合.两个数据集检索的实验结果表明,该方法比其它方法更加有效.     

基于CCA子空间和GMM的自动图像标注

为了避免图像分割,并提高图像标注精度,提出一种基于典型相关分析(CCA)和高斯混合模型(GMM)的自动图像标注方法.利用CCA对图像的全局颜色特征与全局局部二值模式(LBP)纹理特征进行特征融合.使用融合后的语义特征,对每一个关键词建立GMM模型来估计单词类密度,从而在特征子空间中得到每个单词的概率分布.采用贝叶斯分类器确定每个标注词和测试图像的联合概率,运用词间语义关系优化标注结果.实验结果表明,使用该方法后的图像标注性能有了较大程度的改善.     

混合多尺度分析和改进PCNN相结合的图像融合方法

为了更好地融合全色图像中的空间细节信息和多光谱图像中的光谱信息,提出一种基于混合多尺度分析和改进脉冲耦合神经网络(PCNN)的多光谱与全色图像融合方法.首先对全色图像和多光谱图像进行非下采样剪切波变换(NSST),并结合不同多尺度分析方法的互补特性,利用平稳小波变换(SWT)对低频分量部分进行二次分解,在混合多尺度域进行系数融合及SWT逆变换;然后采用基于PCNN的融合规则对高频分量部分进行融合;最后对融合后的高低频系数进行NSST逆变换,得到融合图像.在2组卫星拍摄的多光谱和全色图像上的实验结果表明,在主观视觉与客观评价指标的总体效果上,该方法优于其他8种经典以及流行方法.     

基于脑功能超网络的多特征融合分类方法

针对在超网络上提取局部脑区指标作为特征,忽视了全局的拓扑信息,继而影响网络拓扑的评估,降低分类器性能的问题,提出了一种基于脑功能超网络的多特征融合分类方法,该方法首先在抑郁症数据集上构建超网络,其次将局部脑区特征和子图特征进行融合。最后采用基于多核的SVM分类器进行分类。为了验证所提方法的有效性,选取28例正常被试和38例抑郁症患者进行实验,结果表明,该方法获得了令人满意的分类准确率,平均可达91.60%。获得的异常区域包括左侧舌回、左侧尾状核、左侧丘脑等重要的抑郁症病发区域。故而该基于脑功能超网络的多特征融合分类方法可以有效地用于分类正常人和抑郁症患者。     

基于HOG多特征融合与随机森林的人脸识别

针对人脸识别在复杂环境下识别率低的问题,提出了一种基于梯度直方图(HOG)多特征融合与随机森林的人脸识别方法.该方法通过HOG特征描述子对人脸进行特征提取.首先以网格作为采样窗在整个人脸图上进行整体HOG特征的提取,并将人脸图像分成均匀子块,在包含有人脸关键部分的子块中提取局部HOG特征.然后通过二维主成分分析(2DPCA)和线性判别分析(LDA)对整体和局部特征进行降维,并进行特征层融合形成最终分类特征,最后通过随机森林分类器对其进行分类.FERET人脸库、CAS-PEAL-R1人脸库、真实场景人脸库实验表明,该方法对光照具有鲁棒性,且有较高的识别率和较短的识别时间.     

非下采样Contourlet变换耦合区域信息特征的遥感图像融合算法

当前遥感图像融合算法主要是通过图像的能量信息来完成低频系数的融合,忽略了图像的光谱信息特征,导致融合图像中存在光谱扭曲等不足。设计基于非下采样Contourlet变换与区域信息特征的遥感图像融合算法。引入HSV(Hue,Saturation,Value)变换,从多光谱图像中提取亮度分量。采用非下采样Contourlet变换,对全色图像与多光谱图像的亮度分量进行分解,获取图像的低频系数与高频系数。联合低频系数的区域能量以及信息熵特征,构造低频系数的融合模型,完成低频信息的融合。通过高频系数的区域方差相似度,建立高频系数融合规则,对高频系数完成融合。通过非下采样Contourlet逆变换与HSV逆变换,获取融合图像。实验结果表明,与当前遥感图像融合方法相比,该算法的融合图像具有更好的光谱与空间特性。     

基于NSCT和自适应PCNN的遥感图像融合方法

以全色和多光谱遥感图像为研究对象,提出一种基于非下采样Contourlet变换(NSCT)和自适应脉冲耦合神经网络(PC-NN)的遥感图像融合方法;该方法首先对全色图像和进行过IHS变换的多光谱图像的亮度分量进行NSCT变换,得到低频子带系数和各带通子带系数;其次对低频子带系数采取一种基于边缘的方法以得到融合图像的低频子带系数;然后采用以各带通子带系数的梯度作为PCNN的链接强度β的PCNN图像融合方法来确定融合图像的各带通子带系数;最后经过NSCT逆变换和IHS逆变换得到融合图像;实验结果表明,此方法更好地保留了原遥感图像中的有用信息,并提高了融合图像的质量。     

基于非下采样Contourlet变换和区域特征的医学图像融合

针对非下采样Contourlet变换具有多尺度分析及平移不变的性质,结合计算机断层成像(CT)和核磁共振(MRI)医学图像各自的成像特性,提出了基于非下采样Contourlet变换和区域特征策略来对低频、高频子带进行融合的医学图像融合方法;介绍了图像融合的评价标准,阐述了非下采样Contourlet变换的原理及实现;从视觉效果和客观数据指标方面对融合图像进行主观评判和数值评价。下颌骨系统CT和MRI图像的融合实验结果表明,该方法相对于小波变换和Contourlet变换方法,可有效综合这两种断层图像的有效信息和细节信息,融合后图像具有更优的视觉质量和量化指标。     

人眼视觉感知特性的非下采样Contourlet变换域多聚焦图像融合

目的针对传统的单特征融合方法不足以衡量像素清晰度的问题,同时综合考虑非下采样Contourlet变换(NSCT)系数特点及人眼视觉感知特性,提出一种基于NSCT的多聚焦图像融合方法。方法首先对来自同一场景待融合的源图像进行NSCT变换;然后对低频分量采用基于局部可见度、局部视觉特征对比度和局部纹理特征的综合特征信息进行融合;对高频分量采用基于邻域和兄弟信息归一化的关联权重局部视觉特征对比度进行融合;最后进行逆NSCT变换得到融合图像。结果将本文方法与传统离散小波变换(DWT)、移不变小波变换(SIDWT)、CT(Contottral变换)、NSCT及基于邻域和兄弟信息的NSCT域多聚焦图像融合方法进行了实验对比,本文方法能获得更好的视觉效果以及较大的边缘信息保留值和互信息值。结论定性和定量的实验结果表明了本文方法的有效性。     

基于模糊积分的不完全小波包子空间集成人脸识别

提出一种基于模糊积分的不完全小波包子空间集成人脸识别方法,并与五种相关方法进行实验比较.首先对人脸图像做不完全小波包分解,对双向低频子空间图像直接进行特征提取,对含有一个方向低频成分的高频子空间图像先求平均,再进行提取特征;然后用得到的不同子空间图像训练模糊分类器;最后用模糊积分融合训练的模糊分类器.该方法能够充分利用不同频率小波子空间图像中包含的有用信息,从而提高人脸识别的精度.在ORL、YALE、JAFFE和FERET这4个人脸数据库上进行实验,实验结果表明该方法在识别精度方面均优于五种相关方法.     

多特征融合的道路车辆检测方法

通过改进基于Haar-like特征和Adaboost的级联分类器,提出一种融合Haar-like特征和HOG特征的道路车辆检测方法。在传统级联分类器的Harr-like特征基础上引入HOG特征;为Haar-like特征和HOG特征分别设计不同形式的弱分类器,对每一个特征进行弱分类器的训练,用Gentle Adaboost算法代替Discrete Adaboost算法进行强分类器的训练;在级联分类器的最后几层上使用Adaboost算法挑选出来的特征组成特征向量训练SVM分类器。实验结果表明所提出的方法能有效检测道路车辆。     

Fusion of covariance matrices of PCA and FLD

In this paper, we propose a novel approach for fusing two classifiers, specifically classifiers based on subspace analysis, during feature extraction. A method of combining the covariance matrices of the Principal Component Analysis (PCA) and Fisher Linear Discriminant (FLD) is presented. Unlike other existing fusion strategies which fuse classifiers either at data level, or at feature level or at decision level, the proposed work combines two classifiers while extracting features introducing a new unexplored area for further research. The covariance matrices of PCA and FLD are combined using a product rule to preserve the natures of both covariance matrices with an expectation to have an increased performance. In order to show the effectiveness of the proposed fusion method, we have conducted a visual simulation on iris data. The proposed model has also been tested by performing clustering on standard datasets such as Zoo, Wine, and Iris. To study the versatility of the proposed method we have carried out an experimentation on sports video shot retrieval problem. The experimental results signify that the proposed fusing approach has an improved performance over individual classifiers.     

An investigation of the effects of correlation and autocorrelation on classifier fusion and optimal classifier ensembles

The purpose of this research was to study various fusion strategies where the levels of correlation between features and auto-correlation within features could be controlled. The fusion strategies were chosen to reflect decision-level fusion (ISOC and ROC), feature level fusion, via a single Generalized Regression Neural Network (GRNN) employing all available features, and an intermediate level of fusion that employed the outputs of individual classifiers, in this case posterior probability estimates, before they are subjected to thresholds and mapped into decisions. This latter scheme involved fusing the posterior probability estimates by employing them as features in a probabilistic neural network. Correlation was injected into the data set both within a feature set (auto-correlation) and across feature sets, and sample size was varied for a two class problem. The fusion methods were then extended to three classifiers, and a method is demonstrated that selects the optimal classifier ensemble.     

Mask R-CNN中特征不平衡问题的全局信息融合方法

特征不平衡问题是影响神经网络检测效率的关键因素。针对Mask R-CNN中的特征不平衡问题,提出一种基于全局特征金字塔网络（GFPN）的信息融合方法。通过将GFPN产生的不同大小特征相融合,生成包含全局语义信息的特征网络,并采用反向过程对原始特征层进行重新标度,从而使得每个特征层均含有全局语义信息。实验结果表明,与原始基于Mask R-CNN的方法相比,该方法的检测精度提升4~6个百分点,而检测时间仅增加0.112 s。     

A comparative study on feature extraction for fingerprint classification and performance improvements using rank-level fusion

Fingerprint classification represents an important preprocessing step in fingerprint identification, which can be very helpful in reducing the cost of searching large fingerprint databases. Over the past years, several different approaches have been proposed for extracting distinguishable features and improving classification performance. In this paper, we present a comparative study involving four different feature extraction methods for fingerprint classification and propose a rank-based fusion scheme for improving classification performance. Specifically, we have compared two well-known feature extraction methods based on orientation maps (OMs) and Gabor filters with two new methods based on "minutiae maps" and "orientation collinearity". Each feature extraction method was compared with each other using the NIST-4 database in terms of accuracy and time. Moreover, we have investigated the issue of improving classification performance using rank-level fusion. When evaluating each feature extraction method individually, OMs performed the best. Gabor features fell behind OMs mainly because their computation is sensitive to errors in localizing the registration point. When fusing the rankings of different classifiers, we found that combinations involving OMs improve performance, demonstrating the importance of orientation information for classification purposes. Overall, the best classification results were obtained by fusing orientation map with orientation collinearity classifiers.     

Feature selection with dynamic mutual information

Feature selection plays an important role in data mining and pattern recognition, especially for large scale data. During past years, various metrics have been proposed to measure the relevance between different features. Since mutual information is nonlinear and can effectively represent the dependencies of features, it is one of widely used measurements in feature selection. Just owing to these, many promising feature selection algorithms based on mutual information with different parameters have been developed. In this paper, at first a general criterion function about mutual information in feature selector is introduced, which can bring most information measurements in previous algorithms together. In traditional selectors, mutual information is estimated on the whole sampling space. This, however, cannot exactly represent the relevance among features. To cope with this problem, the second purpose of this paper is to propose a new feature selection algorithm based on dynamic mutual information, which is only estimated on unlabeled instances. To verify the effectiveness of our method, several experiments are carried out on sixteen UCI datasets using four typical classifiers. The experimental results indicate that our algorithm achieved better results than other methods in most cases.     

一种KCCA融合局部特征和全局特征的目标识别算法

提出一种新的KCCA特征融合算法。首先分别提取目标图像的局部特征SIFT和全局Pseudo-Zernike矩特征,并利用K-means算法对局部特征进行预处理;然后利用KCCA将两种特征提取相关特征进行融合,最后将融合特征送入SVM分类器。对遥感飞机图像库做了分类识别的仿真实验。相比于单一特征和CCA特征融合的识别策略,KCCA识别率得到明显提高,理论分析和实验结果证实了该算法具有良好的准确性与可靠性,能够有效提高图像分类识别系统的准确度。     

基于特征融合和混类增强的深度学习滚动轴承故障诊断

轴承故障诊断在维护旋转机械设备和规避重大灾难事故等方面起着至关重要的作用. 针对现有故障诊断模型无法适应实际工业应用中变化的工作负载的问题, 提出了一种基于特征融合和混类增强的故障诊断方法. 首先, 在原始信号的基础上融合时频特征、工况特征和时间差分特征形成新的特征信号; 然后, 采用相空间重构理论将信号特征转换为图像信号, 在训练时通过混类增强拓展数据的分布; 最后, 利用残差网络进行故障诊断分析. 在CWRU数据集上的实验结果表明, 该方法在同工况下的预测精度高达100%, 在变工况下的平均预测精度高达93.28%, 域适应性强.     

一种基于局部排序PCA的线性鉴别算法

主分量分析(Principal Component Analysis,PCA)是模式识别领域中一种重要的特征抽取方法,该方法通过K-L展开式来抽取样本的主要特征。基于此,提出一种拓展的PCA人脸识别方法,即分块排序PCA人脸识别方法(MSPCA)。分块排序PCA方法先对图像矩阵进行分块,对所有分块得到的子图像矩阵利用PCA方法求出矩阵的所有特征值所对应的特征向量并加以标识;然后找出这些所有的特征值中k个最大的特征值所对应的特征向量,用这些特征向量分别去抽取所属的子图像的特征;最后,在MSPCA的基础上,将抽取子图像所得到的特征矩阵合并,把这个合并后的特征矩阵作为新的样本进行PCA+LDA。与PCA和PCA+LDA方法相比,分块排序PCA由于使用子图像矩阵,可以避免使用奇异值分解理论,从而更加简便。在ORL人脸库上的实验结果表明,所提出的方法在识别性能上明显优于经典的PCA和PCA+LDA方法。     

Feature fusion for basic behavior unit segmentation from video sequences

It has become increasingly popular to study animal behaviors with the assistance of video recordings. An automated video processing and behavior analysis system is desired to replace the traditional manual annotation. We propose a framework for automatic video based behavior analysis systems, which consists of four major modules: behavior modeling, feature extraction from video sequences, basic behavior unit (BBU) discovery and complex behavior recognition. BBU discovery is performed based on features extracted from video sequences, hence the fusion of multiple dimensional features is very important. In this paper, we explore the application of feature fusion techniques to BBU discovery with one and multiple cameras. We applied the vector fusion (SBP) method, a multi-variate vector visualization technique, in fusing the features obtained from a single camera. This technique reduces the multiple dimensional data into two dimensional (SBP) space, and the spatial and temporal analysis in SBP space can help discover the underlying data groups. Then we present a simple feature fusion technique for BBU discovery from multiple cameras with the affinity graph method. Finally, we present encouraging results on a physical system and a synthetic mouse-in-a-cage scenario from one, two, and three cameras. The feature fusion methods in this paper are simple yet effective.