一种基于SIFT特征的铁道检测图片的匹配方法

提出一种基于SIFT特征的铁道检测图片的匹配方法。由人工标定铁路上的目标位置图片,通过匹配算法计算匹配图片与目标位置图片可匹配SIFT特征点的数量,利用DTW最优路径规划得到全局最优的匹配结果,从匹配结果中得到一张匹配度最高的图片并将其输出用于道路检测。实验证明,该算法在效率和准确度上均有较好的表现。     

基于特征选择的实体关系抽取*

提出了一种实体关系抽取方案,该方案针对实体关系抽取中特征空间维数过高问题,引入了文本分类中的特征选择算法,如信息增益、期望交叉熵和x2统计,实现了特征空间降维。实验结果表明,各特征选择算法均能在尽量保证抽取性能的同时有效地降低向量空间维数,提高分类效率,其中x2统计取得的效果最好。     

一种高噪声低照度图像的隐藏特征修复算法

当前的图像隐藏特征修复算法的特征融合过程为单次迭代,且为根据光照情况建立暗适应函数,导致低照度图像隐藏特征修复结果存在失真问题,图像噪声也偏高。为此,提出一种基于交替优化的低照度图像隐藏特征修复算法。模拟低光照对图像环境的自动应变能力,并根据光照情况设计暗适应函数,对隐藏特征像素点实现边缘拉伸及中值滤波操作,提取处理后的隐藏特征分量数值,建立非线性映射函数,交替优化融合特征信息,实现低照度图像隐藏特征的修复。仿真结果证明,所提方法可以有效提高色彩饱和度,并且不易出现失真、特征丢失以及噪声现象,在最大程度上保证原始图像的自身特征属性,实现合理有效的隐藏特征修复。     

An end-to-end neural network for detecting hidden people in images based on multiple attention network

Camouflaged people like soldiers on the battlefield or even camouflaged objects in the natural environments are hard to be detected because of the strong resemblances between the hidden target and the background. That’s why seeing these hidden objects is a challenging task. Due to the nature of hidden objects, identifying them require a significant level of visual perception. To overcome this problem, we present a new end-to-end framework via a multi-level attention network in this paper. We design a novel inception module to extract multi-scale receptive fields features aiming at enhancing feature representation. Furthermore, we use a dense feature pyramid taking advantage of multi-scale semantic features. At last, to locate and distinguish the camouflaged target better from the background, we develop a multi-attention module that generates more discriminative feature representation and combines semantic information with spatial information from different levels. Experiments on the camouflaged people dataset show that our approach outperformed all state-of-the-art methods.

     

Relevance-based feature extraction for hyperspectral images.

Hyperspectral imagery affords researchers all discriminating details needed for fine delineation of many material classes. This delineation is essential for scientific research ranging from geologic to environmental impact studies. In a data mining scenario, one cannot blindly discard information because it can destroy discovery potential. In a supervised classification scenario, however, the preselection of classes presents one with an opportunity to extract a reduced set of meaningful features without degrading classification performance. Given the complex correlations found in hyperspectral data and the potentially large number of classes, meaningful feature extraction is a difficult task. We turn to the recent neural paradigm of generalized relevance learning vector quantization (GRLVQ) [B. Hammer and T. Villmann, Neural Networks vol. 15, pp. 1059-1068, 2002], which is based on, and substantially extends, learning vector quantization (LVQ) [T. Kohonen, Self-Organizing Maps, Berlin, Germany: Springer-Verlag, 2001] by learning relevant input dimensions while incorporating classification accuracy in the cost function. By addressing deficiencies in GRLVQ, we produce an improved version, GRLVQI, which is an effective analysis tool for high-dimensional data such as remotely sensed hyperspectral data. With an independent classifier, we show that the spectral features deemed relevant by our improved GRLVQI result in a better classification for a predefined set of surface materials than using all available spectral channels.     

新的纹理图像特征提取方法

提出了一种新的基于非下采样Contourlet变换的纹理特征提取方法.首先对纹理图像进行非下采样Contourlet变换,然后提取不同尺度、不同方向上变换系数矩阵的均值和方差作为特征向量,大大降低了特征维数,并利用BP神经网络进行训练和仿真,实现了纹理图像的自动分类.实验结果表明,与小波包变换和改进的LBP纹理算子等方法相比,该方法能取得更好的分类效果.     

基于眼底图像不同彩色通道的出血特征提取

针对眼底出血图像中出血形态各异、干扰目标多的特性,为提高出血检测精度,同时降低非出血目标引起的干扰,提出了一种基于眼底图像三个彩色通道的出血特征提取方法。该方法利用眼底出血图像在不同彩色通道的表现特性,统计和分析相关性状的像素值特性,并依据出血部分的统计特性设定提取阈值提取出血;使用多尺度顶帽变换和血管密度特征定位血管和黄斑;最后利用不用图像间的逻辑关系针对性去除血管、黄斑干扰,实现了出血区域的自动提取和干扰目标的排除。仿真结果表明,所提方法能够相对完整和准确地提取眼底图像出血目标,且时间效率高。     

Driver identification based on hidden feature extraction by using adaptive nonnegativity-constrained autoencoder

In this paper, we propose a new driver identification method using deep learning. Existing driver identification methods have the disadvantages that the size of the sliding time window is too large and the feature extraction is relatively subjective, which leads to low identification accuracy and long prediction time. We first propose using an unsupervised three-layer nonnegativity-constrained autoencoder to adaptive search the optimal size of the sliding window, then construct a deep nonnegativity-constrained autoencoder network to automatically extract hidden features of driving behavior to further complete driver identification. The results from the public driving behavior dataset indicate that relative to conventional sparse autoencoder, dropout-autoencoder, random tree, and random forest algorithms, our method can effectively search the optimal size of the sliding time window, and the window size is shortened from the traditional 60s to 30s, which can better preserve the intrinsic information of the data while greatly reducing the data volume. Furthermore, our method can extract more distinctive hidden features that aid the classifier to map out the separating boundaries among the classes more easily. Finally, our method can significantly shorten the prediction time and improve the timeliness under the premise of improving the driver identification performance and reducing the model overfitting.     

数据挖掘提取查询树特征的SQL注入攻击检测

针对Web应用中数据库信息容易遭受SQL注入攻击的问题,提出一种基于数据挖掘技术的SQL注入攻击检测方法,其核心在于查询树特征的提取和转换。首先,在SQL数据库日志中收集内部查询树;然后,提取查询树中的语义和语法特征,并通过利用多维序列作为中间表示将查询树特征转换为一个n维字符特征向量;再后,根据查询树类型,利用不同的统计模型将字符特征向量转换成n维数值特征向量;最后,根据这些特征,利用多项式核函数SVM对其进行分类,从而实现SQL攻击检测。实验结果表明,相比其他几种较新的方案,提出的方案有效提高了SQL攻击的正确检测率。     

适用于倾斜影像的加速KAZE-SIFT特征提取算法

针对传统正摄影像的特征提取算法处理倾斜影像匹配效果不佳的问题，在已有特征提取算法的基础上，提出了一种适用于倾斜影像的特征提取算法--加速KAZE-尺度不变特征变换（AKAZE-SIFT）算法。首先，为保证特征检测的准确性与独特性，采用充分保留图像轮廓信息的加速KAZE（AKAZE）算子进行特征检测；其次，为提升特征描述的稳定性，采用稳健的尺度不变特征变换（SIFT）算子进行特征描述；然后，依据目标特征向量和候选特征向量间的欧氏距离确定粗匹配点对；最后，采用随机抽样一致性算法进行单应性约束，提高匹配纯度。模拟影像在倾斜摄影条件下的模糊、旋转、亮度、视角和尺度变化，对特征提取算法性能进行评估，实验结果表明，AKAZE-SIFT算法相比SIFT算法和AKAZE算法召回率分别提高了12.8%和5.3%，精准率提高了6.5%和6.1%，F1值提升了13.8%和5.6%；提取效率优于SIFT算法，略逊于AKAZE。AKAZE-SIFT算法具有良好的检测和描述能力，更适用于倾斜影像特征提取。     

A CUDA-powered method for the feature extraction and unsupervised analysis of medical images

The Journal of Supercomputing - Image texture extraction and analysis are fundamental steps in computer vision. In particular, considering the biomedical field, quantitative imaging methods are...     

基于偏度和结构特征的无参考虚拟视点图像质量评价

针对虚拟视点图像存在与普通图像不同失真的问题,提出了一种基于偏度和结构特征的无参考虚拟视点图像质量评价方法.首先,将输入图像分为H、S、V三个通道,每个通道都均分为九个小块,对每个小块提取偏度特征;然后,利用局部二值模式算子对结构特征映射进行编码,计算质量感知分数;最后,将偏度特征和结构特征输入支持向量机进行训练,得到无参考的虚拟视点图像回归与视觉质量预测模型来预测图像质量.在IRCCyN/IVC和MCL-3D两个公共虚拟视点图像数据库上对提出方法进行了实验.实验结果表明,提出方法的皮尔逊线性相关系数(PLCC)分别为0.8538和0.9534,斯皮尔曼秩相关系数(SROCC)分别为0.7966和0.9159.提出方法的PLCC和SROCC均高于自回归加阈值(APT)等10个虚拟视点图像质量评价方法和BRISQUE等6个通用的无参考质量评价方法.该方法采用的偏度和结构特征能很好地评价虚拟视点图像的视觉质量,评价结果与主观感知有较好的一致性.     

基于偏度和结构特征的无参考虚拟视点图像质量评价

针对虚拟视点图像存在与普通图像不同失真的问题,提出了一种基于偏度和结构特征的无参考虚拟视点图像质量评价方法.首先,将输入图像分为H、S、V三个通道,每个通道都均分为九个小块,对每个小块提取偏度特征;然后,利用局部二值模式算子对结构特征映射进行编码,计算质量感知分数;最后,将偏度特征和结构特征输入支持向量机进行训练,得到无参考的虚拟视点图像回归与视觉质量预测模型来预测图像质量.在IRCCyN/IVC和MCL-3D两个公共虚拟视点图像数据库上对提出方法进行了实验.实验结果表明,提出方法的皮尔逊线性相关系数(PLCC)分别为0.8538和0.9534,斯皮尔曼秩相关系数(SROCC)分别为0.7966和0.9159.提出方法的PLCC和SROCC均高于自回归加阈值(APT)等10个虚拟视点图像质量评价方法和BRISQUE等6个通用的无参考质量评价方法.该方法采用的偏度和结构特征能很好地评价虚拟视点图像的视觉质量,评价结果与主观感知有较好的一致性.     

面向RGB-D图像的多层特征提取算法综述

RGB-D图像包含丰富的多层特征,如底层的线特征、平面特征,高层的语义特征,面向RGB-D图像的多层特征提取结果可以作为先验知识提升室内场景重建、SLAM（simultaneous localization and mapping）等多种任务的输出质量,是计算机图形学领域的热点研究内容之一。传统的多层特征提取算法一般利用RGB图像中丰富的颜色、纹理信息以及深度图像中的几何信息提取多层特征,此类提取算法依赖输入RGB-D图像的质量,而受采集过程中环境和人为因素的影响,很难得到高质量的RGB-D图像。随着深度学习技术的快速发展,基于深度学习的多层特征提取算法突破了这一限制,涌现出一批高质量的研究成果。本文对面向RGB-D图像的多层特征提取算法进行综述。首先,汇总了现有的常用于多层特征提取任务的RGB-D数据集和相关算法的质量评价指标。然后,按照特征所处的不同层次,依次对线、平面和语义特征相关算法进行了总结。此外,本文还对各算法的优缺点进行比较并结合常用算法质量评价标准进行了定量分析。最后,讨论了当前多层特征提取算法亟待解决的问题并展望了未来发展的趋势。     

Efficient relation extraction method based on spatial feature using ELM

Entity relation extraction can be applied in the automatic question answering system, digital library and many other fields. However, the previous works on this topic mainly focused on the features from a sentence itself in the data sets, without considering the links between sentences in the corpus. In this paper, we propose a concept model and obtain a new effective spatial feature based on this concept model. The added feature makes our feature space concerning not only the inherent information of the sentence itself, but also the semantic information connection between sentences. At last, we use ELM as the training classifier in entity relation extraction. The experiment result shows that the precision and recall of the relation extraction both have a significant increase, by using the new feature. Also, the use of ELM significantly reduces the time of relation extraction. It has a better performance than the traditional method based on SVM.     

基于边缘几何不变性的特征提取算法研究

针对异源图像中可见光与红外图像特征提取算法进行研究,提出了一种基于边缘几何不变性特征的提取算法。采用空域滤波及灰度处理的方法对背景噪声进行处理,使用C anny算子提取目标的边缘,利用二值形态学中腐蚀与膨胀两种基本运算对边缘细化填充。利用圆形模板匹配的方法提取边缘特征点,通过对有效特征点的筛选与组合形成不同的三角形区域特征,并计算这些三角形区域的几何特性。仿真实验结果表明,该方法可以有效提取异源图像的共有特征。     

基于新型特征提取的寄生虫卵图像识别研究

讨论了用支持向量机进行多分类的若干学习策略,提出了一种新型图像特征提取方法,以此来实现对鞭虫等九种寄生虫卵图像自动识别和分类,平均识别率优于传统神经网络,达到了93.9%,为寄生虫卵图像识别提供了一种新方法。     

Spectral reflectance of coral reef bottom-types worldwide and implications for coral reef remote sensing

Coral reef benthic communities are mosaics of individual bottom-types that are distinguished by their taxonomic composition and functional roles in the ecosystem. Knowledge of community structure is essential to understanding many reef processes. To develop techniques for identification and mapping of reef bottom-types using remote sensing, we measured 13,100 in situ optical reflectance spectra (400-700 nm, 1-nm intervals) of 12 basic reef bottom-types in the Atlantic, Pacific, and Indian Oceans: fleshy (1) brown, (2) green, and (3) red algae; non-fleshy (4) encrusting calcareous and (5) turf algae; (6) bleached, (7) blue, and (8) brown hermatypic coral; (9) soft/gorgonian coral; (10) seagrass; (11) terrigenous mud; and (12) carbonate sand. Each bottom-type exhibits characteristic spectral reflectance features that are conservative across biogeographic regions. Most notable are the brightness of carbonate sand and local extrema near 570 nm in blue (minimum) and brown (maximum) corals. Classification function analyses for the 12 bottom-types achieve mean accuracies of 83%, 76%, and 71% for full-spectrum data (301-wavelength), 52-wavelength, and 14-wavelength subsets, respectively. The distinguishing spectral features for the 12 bottom-types exist in well-defined, narrow (10-20 nm) wavelength ranges and are ubiquitous throughout the world. We reason that spectral reflectance features arise primarily as a result of spectral absorption processes. Radiative transfer modeling shows that in typically clear coral reef waters, dark substrates such as corals have a depth-of-detection limit on the order of 10-20 m. Our results provide the foundation for design of a sensor with the purpose of assessing the global status of coral reefs.