Building Detection Using Local Gabor Features in Very High Resolution Satellite Images
- 格式:pdf
- 大小:1.18 MB
- 文档页数:4
基于连通区域和统计特征的图像文本定位刘亚亚;于凤芹;陈莹【摘要】文本定位是图像中文本提取的前提与基础.针对场景图像中背景复杂和光照影响,提出一种由粗略到精确的文本定位算法.该算法首先在边缘图像上利用连通区域分析进行粗略定位得到文本候选区域,然后提取候选区域的方向梯度直方图特征和改进的局部二值模式特征进行分类,去除虚假文本达到精确定位.仿真实验结果表明,该算法能够有效地降低背景复杂与光照不均的影响,在场景图像中准确地定位文本区域.【期刊名称】《计算机工程与应用》【年(卷),期】2016(052)005【总页数】5页(P165-168,208)【关键词】文本定位;连通区域分析;方向梯度直方图特征;局部二值模式特征【作者】刘亚亚;于凤芹;陈莹【作者单位】江南大学物联网工程学院,江苏无锡214122;江南大学物联网工程学院,江苏无锡214122;江南大学物联网工程学院,江苏无锡214122【正文语种】中文【中图分类】TN911.73图像中文本信息是描绘和理解图像内容的重要信息,文本区域的定位是文本提取非常重要的步骤与基础,准确的文本区域的定位才能保证文本信息提取的有效性。
然而,由于背景复杂、光照变换、字体大小和方向的多变等原因,自然场景图像中的文本定位具有更多的不确定性和难度,是目前研究的难点。
文本定位的方法通常分为基于连通区域、基于边缘检测和基于纹理特征的三类算法[1]。
基于连通区域的算法是利用图像中的文本颜色相似并与背景颜色相差较大的特征进行文本定位的,但是对光照和颜色比较敏感,对背景复杂的图像效果不理想;Pan等[2]设计一个文本区域探测器生成文本置信图,然后利用条件随机域模型进行连通域分析,得到文本区域;Shivakumara等[3]首先通过傅里叶-拉普拉斯变换对图像进行滤波,然后基于最大差值用K-means聚类得到文本区域,可检测非水平方向上的文本;Hinnerk Becker等[4]首先采用一种自适应二值化的算法在图像中提取字母,然后利用几何约束的方法将字母连接成文本行。
基于Log-Gabor滤波和LBP算子的光照不变人脸识别方法程雪峰;李顺;龙飞【摘要】提出了一种基于Log Gabor滤波和局部二值模式(local binary patterns,LBP)算子的光照不变人脸识别方法.该方法首先对人脸图像进行对数变换预处理,有效改善剧烈光照变化对人脸图像的不利影响.然后采用Log Gabor滤波器与图像进行卷积,得到不同尺度和不同方向下的人脸Log Gabor特征图像.在此基础上,再使用LBP算子对Log Gabor图像进行描述,最后将所有的Log Gabor图像的LBP特征进行简单连接,作为人脸的特征向量.将所提出的方法在YaleB数据库上进行实验,实验结果表明该方法能够有效提高复杂光照条件下的人脸识别率.【期刊名称】《厦门大学学报(自然科学版)》【年(卷),期】2014(053)003【总页数】5页(P359-363)【关键词】人脸识别;局部二值模式;Log-Gabor【作者】程雪峰;李顺;龙飞【作者单位】厦门大学软件学院,福建厦门361005;厦门大学软件学院,福建厦门361005;厦门大学软件学院,福建厦门361005【正文语种】中文【中图分类】TP391.41在人脸识别所面临的挑战中,光照问题一直是人们研究的重点和难点.由于光照的不同,相同个体人脸图像之间的差异要大于由不同人脸带来的差异.因此,光照问题一直是影响人脸识别系统性能的一个十分关键的因素[1-2].为了克服光照的变化给人脸识别带来的不利影响,学者们针对这个问题做了大量的研究.光照不变特征提取方法是解决光照问题的重要方法之一,该方法的核心思想是通过提取人脸图像对光照不变或者不敏感的特征用于人脸识别,从而达到降低光照变化给识别所带来的不利影响.这类方法在解决光照问题中得到了广泛的应用,国内外学者对此做了大量的研究,提出了很多光照不变特征提取的方法.如Belhumeur等[3]提出的比率图方法,通过比较2幅图像的比率图像的复杂度,来判断2幅图像是来自于不同光照下的同一物体,还是来自于不同的2个物体.Sawides等[4]提出的相位图方法用于光照不变人脸识别,其基本思想是将图像从空域经快速傅里叶变换(FFT)变换到频域后,相位图比能量图含有更多的对识别有用的信息,且相位图对光照变化和遮挡较不敏感.Wang 等[5]提出的自商图像(self quotient image)法,利用在Retinex理论的基础上提出的自商图像用于人脸识别,实验在卡内基梅隆大学人脸表情(CMU-PIE)数据库上获得了较好的效果.Chen等[6]提出了LOG-DCT(离散余弦变换)方法用于光照不变特征的提取,通过在对数域中进行DCT变换进而提取光照不变特征用于人脸识别.本文提出了一种对光照变化较为鲁棒的人脸识别方法,通过将 Log-Gabor滤波[7-8]和局部二值模式(local binary patterns,LBP)描述相结合提取人脸的光照不变特征.该方法首先对人脸图像进行对数变换从而有效改善剧烈光照变化对人脸图像的不利影响,然后利用Log-Gabor滤波器对人脸图像进行滤波,得到不同尺度、不同方向下的Log-Gabor特征图像,在此基础上再利用LBP算子对每个特征图像进行描述,最后将所有Log-Gabor特征图像的LBP特征连接起来,构成人脸特征向量.我们在YaleB数据库上进行了实验,实验结果表明本文方法能够有效提高复杂光照条件下的人脸识别率.1 Log-Gabor滤波器和LBP算子1.1 Log-Gabor滤波器Field在1987年提出了Log-Gabor函数[7],他通过研究发现,在对数频率尺度下的高斯转移函数能够更好地描述自然图像.Log-Gabor滤波器在频域的定义如下:式中,f0为滤波器中心频率,k为控制带宽.Log-Gabor滤波器有2个显著的特点.首先,其没有直流分量,因此在进行图像处理时受光照条件的影响较小,可以一定程度上克服光照给人脸识别带来的不利影响;其次,Log-Gabor滤波器的转移函数在高频部分有一个很长的延伸,它可以改善普通Gabor滤波器低频表示过度而高频表示不足的缺点.将人脸图像I与4个尺度、6个方向的Log-Gabor滤波器进行卷积,并取其幅度信息(能量)作为输出.人脸图像经Log-Gabor滤波后的结果如图1所示.图1 人脸图像的Log-Gabor特征描述Fig.1 Log-Gabor feature descriptionof face image1.2 LBP算子LBP是一种有效的纹理描述算子[9],具有计算简单、鉴别能力强等优点.基本LBP算子示意图如图2所示.对于一个3×3窗口,若窗口的中心像素点定义为gc,相邻的8个像素的灰度值分别为g0,…,g7,则纹理T可以定义为:其中,然后根据像素的不同位置赋予不同的权重并加权求和,按公式4计算,便可得到该窗口的LBP值.根据以上的操作方法,对图像的每一个像素应用LBP算子,就会得到一个相应的LBP值,并以此来描述图像的纹理信息.图2 基本LBP算子示意图Fig.2 Basic LBP operator2 LG-LBP方法本文在Log-Gabor滤波和LBP算子的基础上,提出了一种新的人脸识别方法,以下简称LG-LBP方法.基本做法是,先对输入的原始人脸图像进行对数变换预处理,以在一定程度上消除光照条件对识别结果的影响.对变换后的人脸图像进行Log-Gabor滤波,得到不同尺度和不同方向下的人脸局部特征图像.为了进一步提高特征的鉴别能力,再分别从每个特征图像中提取LBP特征.最后,将所有特征图像的LBP特征进行连接,构成用于识别的LG-LBP特征向量.识别阶段,采用基于欧式距离的最近邻分类器.本文LG-LBP方法的流程如图3所示.图3 LG-LBP方法流程图Fig.3 Diagram of LG-LBP method2.1 对数变换预处理在利用Log-Gabor滤波器对图像进行滤波之前,我们先对图像进行对数变换,如公式(5)所示:根据Retinex理论所确定的光照模型,对数变换后图像的反射分量和光照分量的乘性关系在对数域中变成了加性关系,这样有利于将图像的反射分量和光照分量进行分离,便于分别进行处理.不同光照条件下的人脸图像的对数变换结果如图4所示.可见,对于光照变化比较剧烈的人脸图像(图4(b)),经过对数变换预处理后图像的外观得到了较明显的改善.图4 不同光照条件下的图像及对数变换后的图像Fig.4 Image after logarithmic transformation under variant illumination(a)正常光照条件下的人脸图像及其对数变换的结果;(b)光照变化比较剧烈的人脸图像及其对数变换的结果. 2.2 LG-LBP特征描述经过了上述的对数变换预处理之后,利用Log-Gabor滤波器对预处理后的人脸图像进行滤波.在本文中我们选取了包含4个尺度和6个方向的滤波器组,因此滤波后一共产生24幅Log-Gabor特征图像.然后,利用LBP算子对这些特征图像进行纹理描述,从而获得人脸图像在4个尺度、6个方向下的LGLBP图像,如图5所示.得到LG-LBP图像后,需要对LG-LBP图像进行分块,再统计每个块的LBP模式直方图.最后,将所有LG-LBP图像和所有分块的直方图进行连接,构成人脸特征向量.3 实验图5 LG-LBP图像Fig.5 LG-LBP image本文选择使用YaleB和扩展的YaleB数据库来进行人脸识别的实验.YaleB数据库中共包含了10个人共640张人脸正面图像,其中每人有64张不同光照条件下的图像,而扩展的YaleB数据则包含了28个人共16 128张人脸图像,其光照条件的变化与YaleB相同,但扩展的YaleB中还包含了姿态的变化[11].为了更客观地研究算法是否对光照变化有效,本文只考虑那些存在光照变化的正面图像,因此只选择了扩展的YaleB数据库中每人64张存在不同光照变化的正面人脸图像与YaleB进行合并,除去18张损坏的图像,合并后的数据库中包含38个人共2 421张人脸图像.根据照射光源与相机坐标的夹角的不同将合并后的数据库分成了5个子集,如表1所示.合并后的数据库中的图片都是经过人眼定位和裁剪过的正面图像,大小为192×168,如图6所示.实验中,有以下参数需要设置:1)Log -Gabor滤波器的尺度与方向;2)LBP算子邻域大小;3)LBP图像分块数目.本实验的参数设置如表2所示.表1 合并后的数据库分成的5个子集Tab.1 5subsets of combined facedatabase子集光照角度/(°)子集图像数1 0~12 263 2 13~25 456 3 26~50 455 4 51~77 526 5>77 7143.1 实验1将本文提出的LG-LBP方法与原始的LBP、Log-DCT[6]、Li等提出的局部图像描述方法(LDR)[12]、基于DoG(Difference of Gaussian)滤波的LBP (以下简称 DoG-LBP方法)方法[13]进行了比较.Log-DCT 方法实验结果来自文献[6].由于子集1只有微弱的光照变化,各种方法在子集1上的识别率几乎均为100%,因此,本文只列出子集2~5上的实验结果,如表3所示.由表3可知,LG-LBP方法的平均识别率为99.44%,在全部4个子集上,LG-LBP方法均获得了最好的识别结果.Log-DCT方法在频域中通过设置阈值对低频成分进行了直接去除以达到去除光照成分保留光照不变特征的目的,但在光照严重不足情况下,低频成分中那部分对识别有用的信息也同时被去除,该方法的识别效果也因此受到影响;与DoG-LBP方法相比,本文的方法在光照情况最差的子集5上识别率更高.以上实验表明,本文提出的LG-LBP方法对光照的变化具有较高的鲁棒性.图6 合并后的YaleB数据库5个子集的图像Fig.6 5subset image of YaleB face database表2 参数设置Tab.2 Parameter setting名称参数参数值Log-Gabor滤波器尺度 4方向 6 Uniform LBP描述邻域大小 LBP8,1 LBP图像分块尺寸13×13表3 不同方法识别性能的比较Tab.3 Recognition performance comparison of different methods方法识别率/%子集2 子集3 子集4 子集5 平均值原始LBP 99.60 96.90 62.20 17.23 67.00 Log-DCT[6] 100.00 89.50 89.20 87.40 91.50 LDR[12] 99.56 92.31 88.40 75.91 89.05 DoG-LBP[13] 99.78 99.34 98.67 87.11 96.23 LG-LBP 100.00 100.00 99.43 98.32 99.443.2 实验2在对图像进行滤波前采用对数变换对图像进行预处理能有效提升光照剧烈变化条件下人脸识别的识别率.下面通过在YaleB和扩展的YaleB合并的数据库上的实验进行验证和说明.实验在两种情况下分别计算在合并的YaleB上4个子集的识别率,分别是在Log-Gabor滤波前采用和不采用对数变换.实验结果如表4所示.从实验结果可以发现,在光照条件比较好的子集2和子集3上,不对图像进行对数变换也能够获得比较好的识别效果,而在光照变化比较剧烈的子集4和子集5上,采用对数变换后识别结果明显得到改善.因此,对图像进行对数变换不会影响光照正常情况下有效特征的提取,而在光照变化剧烈的情况下却能提升特征的有效性. 表4 对数变换预处理对识别性能的影响Tab.4 Effects of logarithm transformation on recognition performance识别率/%方法子集2 子集3 子集4 子集5 平均值采用对数变换 100.00 100.00 99.05 98.46 99.38不采用对数变换 100.00 99.78 84.22 77.17 90.293.3 实验3本文在人脸Log-Gabor特征描述的基础上,引入LBP算子得到人脸特征向量.下面通过在合并的YaleB上的实验来分析和讨论引入LBP算子和未引入LBP算子2种情况下的识别效果,实验结果如表5所示.表5 LBP算子在本文方法中的应用Tab.5 The role of LBP descriptor in our method方法识别率/%子集2 子集3 子集4 子集5 平均值Log-Gabor+LBP 100.00 100.00 99.05 98.46 99.38 Log-Gabor 100.00 85.49 53.23 56.44 73.79直接使用人脸的Log-Gabor特征进行识别同样在光照情况比较好的子集2上取得了100%的识别率,然而随着光照情况的恶化,方法的识别效果也急剧下降,在光照变化剧烈的子集4和子集5上只有50%左右的人脸能够被正确的分类和识别.但引入LBP算子后识别的效果得到大幅度的改善,实验结果表明,LBP算子与Log-Gabor方法相结合能够有效提取人脸的光照不变特征.4 结论光照变化一直是困扰人脸识别系统走向实际应用的一个十分关键的问题.本文从提取光照不变特征入手,提出了一种基于Log-Gabor滤波和LBP算子的光照不变人脸识别方法.在著名的YaleB数据库上的实验表明,本文方法能够有效提取对光照较不敏感的人脸特征,尤其是在光照变化剧烈的情况下,本文方法相对于传统方法而言优势更为明显.将该方法应用到表情识别等其他图像识别任务中是我们后续的工作.【相关文献】[1]Zhu J Y,Zheng W S,Lai J H.Logarithm gradient histogram:ageneral illumination invariant descriptor for face recognition[C]∥IEEE Conference on Automatic Face and Gesture Recognition.Shanghai:IEEE,2013:1-8.[2]Zhuang Liansheng,Yang Allen,Zhou Zihan,et al.Singlesample face recognition with image corruption and misalignment via sparse illumination transfer[C]∥C omputer Vision and Pattern Recognition Proceedings.Portland,OR:IEEE,2013:3546-3553. [3]Jacobs D W,Belhumeur P N,Basri paring images under variable illumination [C]∥Computer Vision and Pattern Recognition.Santa Barbara,CA:IEEE,1998:610-617.[4]Sawides M,Kumar B V K V,Khosla P K.Eigenphases vs eigenfaces[C]∥Proceedings of the 17th International Conference on IEEE.[s.l.]:IEEE,2004:810-813.[5]Wang H,Li S Z,Wang Y.Face recognition under varying lighting conditions using self quotient image[C]∥Proceedings Si xth IEEE International Conference on.[S.l.]:IEEE Press,2004:819-824.[6]Chen W,Er M J,Wu S.Illumination compensation and normalization for robust face recognition using discrete cosine transform in logarithm domain[J].Systems,Man,and Cybernetics,Part B:Cybernetics,IEEE Transactions on,2006,36(2):458-466.[7]Field D J.Relations between the statistics of natural images and the response properties of cortical cells[J].J Opt Soc Am A,1987,4(12):2379-2394.[8]Wang Jinglei,Long Fei,Chen Jiping,et al.A novel eye localization method based on Log-Gabor transform and integral image[J].Applied Mathematics & Information Sciences,2012,6:323-329.[9]Ojala T,Pietikäinen M,Harwood D.A comparative study of texture measures with classification based on featured distributions[J].Pattern Recognition,1996,29(1):51-59.[10]Ahonen T,Hadid A,Pietikäinen M.Face recognition with local binary patterns [M].Berlin Heidelberg:Springer,2004:469-481.[11]Georghiades A S,Belhumeur P N,Kriegman D J.From few to many:illumination cone models for face recognition under variable lighting and pose[J].Pattern Analysis and Machine Intelligence,IEEE Transactions on,2001,23(6):643-660.[12]Li S,Long F,Cheng X,et al.Illumination invariant face recognition by local image descriptor in logarithm domain[C]∥Computational Intelligence and Design(ISCID),2012Fifth International Symposium on IEEE.[s.l.]:IEEE,2012:201-204.[13]Lian Z,Er M J,Li J.A novel face recognition approach under illumination variations based on local binary pattern[C]∥ Computer Analysis of I mages and Patterns.Berlin Heidelberg:Springer,2011:89-96.。
摘要在城市以及乡镇的现代化建设不断发展的趋势下,公路、桥梁等基础交通设施的分布范围越来越广。
随着使用时间的增长,这些基础设施由于各种原因(如建材质量、交通工具超载、恶劣天气等)会出现各种病害,如裂缝、下沉、脱空、变形等,容易造成各种交通事故,因此公路、桥梁等基础交通设施的状况调查和护养愈显重要。
同时由于规划与建设的不同步,在布置和建设地下管道、电缆、排水系统等地下设施时,常会遇到与其它工程设施冲突的问题。
因此在正式实施地下设施建设工程前,需要获取地下结构和目标分布等信息,以分析地下工程的可行性。
超宽带探地雷达技术是一种高效、精确的无损探测方式,对浅层目标具有良好的探测效果。
本文叙述了超宽带探地雷达的发展背景,系统组成与技术原理,研究了超宽带探地雷达在浅层目标探测方面的重构与仿真,并提出了一种基于功率谱估计的超宽带探地雷达浅层目标探测方法,同时分别利用RIS-K2探地雷达系统与GprMax2D软件进行实测和仿真实验,在Matlab数值计算环境中对所提出的方法进行目标探测数据处理。
本文的主要研究工作和成果如下:1.对超宽带探地雷达系统和理论进行了研究,对浅层目标进行了模型重构,同时利用基于时域有限差分法(FDTD)的GprMax2D软件对重构模型进行仿真,利用Matlab软件进行目标仿真数据处理。
2.将超宽带探地雷达技术理论应用于具体的实践应用中。
本文使用意大利IDS公司RIS-K2探地雷达系统进行了目标数据采集和目标探测实验。
同时利用Matlab软件对所采集的数据进行了成像和算法处理。
3.本文提出了一种基于功率谱估计的超宽带探地雷达浅层目标探测方法。
该方法主要针对探测深度小于5m的浅层目标探测的应用,减少了探测过程中所需存储的数据量,计算复杂度低,算法处理速度快,可以实现采集过程与数据处理过程的结合。
本文利用RIS-K2探地雷达系统对华南理工大学五山校区内的湖滨北路与嵩山路进行了实测,利用所提方法对探测采集的数据进行了数据处理与数据分析。
优化经验模态模型下的弱信号重构方法文畅;廖虎;谢凯;贺建飚【摘要】为快速恢复强噪声掩盖下的弱信号,提出一种基于ACEEMD(adaptive complete ensemble empirical mode decomposition)模型的弱信号快速重构方法.结合优化的互补集合经验模态分解与GPU并行运算,快速将弱信号从强噪声背景下重构出来,使其更容易被检测.运用该方法对模拟弱信号数据与实际弱信号数据进行处理,实验结果表明,处理后的测试数据信噪比提高了2 db-3 db,处理速度提高了4-5倍,处理效果明显.%To recover the weak signal with strong noise,a weak signal recovery method based on ACEEMD was proposed.The method combined the optimization of CEEMD and GPU parallel operation,and quickly recovered the weak signal from strong noise,making it easier be processed and identified.This method was used to process the simulated weak signal data and the real weak signal data.Experimental results show that the signal to noise ratio is increased by 2 dB to 3 dB,and the processing speed is increased by 4-5 times and the processing effect is obvious.【期刊名称】《计算机工程与设计》【年(卷),期】2018(039)004【总页数】5页(P1111-1115)【关键词】弱信号;强噪声背景;互补集合经验模态分解;图形处理器;并行计算【作者】文畅;廖虎;谢凯;贺建飚【作者单位】长江大学计算机科学学院,湖北荆州 434023;长江大学电子信息学院,湖北荆州 434023;长江大学电子信息学院,湖北荆州 434023;中南大学信息科学与工程学院,湖南长沙 410083【正文语种】中文【中图分类】TP391.40 引言弱信号具有非线性、非平稳性,所以传统的傅里叶变换、短时傅立叶变换并不能达到在时域和频域同时高精度的要求。
改进的HOG和Gabor,LBP性能比较I. Introduction- Background and motivation- Objectives of the study- Contributions of the paperII. Literature Review- Overview of HOG, Gabor, and LBP feature extraction methods - Advantages and limitations of each method- Related works on improving HOG, Gabor, and LBP performance III. Methodology- Description of the proposed improvements to HOG and Gabor - Description of LBP feature extraction method- Feature selection and dimensionality reduction techniquesIV. Experimental Evaluation- Datasets used for evaluation- Comparison of the performance of the proposed HOG and Gabor improvements versus standard HOG, Gabor, and LBP methods- Statistical analysis and interpretation of resultsV. Conclusion and Future Work- Summary of findings- Implications and potential applications of the proposed improvements- Limitations and future directions for researchI. IntroductionThe field of image and object recognition has gained much attention and development over the past few decades. In order toaccurately and efficiently recognize objects in images, feature extraction methods have been developed to extract relevant information from images that enable classification and detection tasks. The HOG (Histogram of Oriented Gradients), Gabor, and LBP (Local Binary Pattern) feature extraction methods are among the most popular techniques for image and object recognition. However, there is always a possibility for further improvement in their performance.The motivation for this study is to enhance the performance of these widely-used feature extraction techniques in order to achieve better accuracy and speed in image classification and detection tasks. In this paper, we propose novel improvements to the HOG and Gabor feature extraction methods and compare their performance with the standard HOG, Gabor, and LBP methods. The contributions of this study include the development and evaluation of improved feature extraction techniques for image and object recognition.The objectives of this study are as follows:1. To review the current state-of-the-art feature extraction methods for image and object recognition.2. To propose novel improvements to the HOG and Gabor feature extraction methods.3. To evaluate the performance of the proposed improvements and compare them with standard HOG, Gabor, and LBP feature extraction methods.4. To provide a critical analysis and interpretation of the results in order to identify strengths and limitations of the proposed improvements.5. To discuss the implications and potential applications of the proposed improvements and provide directions for future research in this area.In the following sections, we will review the literature on HOG, Gabor, and LBP feature extraction methods, discuss the proposed improvements to HOG and Gabor, describe the methodology for evaluation, present experimental results, and conclude with a discussion of the findings and future research directions.II. Literature ReviewA. Histogram of Oriented Gradients (HOG)The HOG feature extraction method was first introduced by Dalal and Triggs in 2005 for object detection and recognition. The basic idea behind HOG is to divide an image into small cells and to compute the gradient orientation and magnitude within each cell. The gradients are then aggregated into a histogram of oriented gradients for each cell, where the orientations are binned into predetermined intervals. The histograms are also normalized to reduce the impact of variations in illumination and contrast. Finally, the histograms are concatenated across neighboring cells to form a feature vector that represents the entire image.Despite its popularity and success in various applications, HOG has some limitations. One of the main limitations is that HOG only considers local gradient orientations within each cell and may not capture global information about the object of interest. Additionally, HOG requires careful parameter tuning in order to achieve optimal performance.B. Gabor FeaturesGabor features are a widely-used technique for texture analysis and recognition. They are based on the Gabor filter, which is a sinusoidal wave modulated by a Gaussian envelope. Gabor filters are sensitive to both the spatial frequency and orientation of an image. Gabor features can be computed by convolving an image with a set of Gabor filters with different orientations and frequencies. The outputs of the convolutions are then squared and summed, and the resulting feature vectors can be used for classification and recognition tasks.While Gabor features have been shown to be effective for texture analysis, they can be computationally expensive due to the large number of filters required. Additionally, Gabor features may not perform well in situations where the object of interest is not a texture. Therefore, additional improvements are needed to enhance the performance of Gabor features for object recognition tasks.C. Local Binary Patterns (LBP)Local Binary Patterns are a feature extraction method that operate on the texture of an image, and represent the spatial relationship between pixels in a given region. LBP encodes the differences between the pixel values of an image and their neighboring pixels into a binary code. The binary codes are then transformed into histograms which capture the distribution of local patterns within the image. LBP features have been successfully applied in a variety of contexts, such as face recognition, texture classification,and object recognition.While LBP features have been shown to be effective for a variety of tasks, they may not be sufficient for complex object recognition tasks where global information is important. Additionally, they can be sensitive to variations in illumination and contrast, and do not capture the underlying structure of an object.In the next section, we propose novel improvements to the HOG and Gabor feature extraction methods to enhance their performance.III. Proposed MethodIn this section, we propose a novel method that enhances the performance of the HOG and Gabor feature extraction methods for object recognition. Our method combines the strengths of both HOG and Gabor features and overcomes their limitations by introducing a new feature extraction framework that leverages the complementary information provided by the two methods.A. Hybrid HOG-Gabor FeaturesOur proposed method, hybrid HOG-Gabor features, takes inspiration from both HOG and Gabor features by extracting different types of features from an input image. Firstly, we apply the HOG feature extraction method to the image, which provides local orientation information about the edges and gradients in the image. We then apply a set of Gabor filters to the image to extract texture information at different scales and orientations.To combine the information provided by HOG and Gabor features,we compute the weighted sum of the feature vectors obtained from both methods. The weights are optimized using a genetic algorithm to maximize the classification accuracy of the resulting feature vector. The final feature vector is then used for classification using a support vector machine (SVM) classifier.B. Experimental ResultsTo evaluate the effectiveness of our proposed method, we conducted experiments on several benchmark datasets, including the CIFAR-10, CIFAR-100, and Caltech-101 datasets. We compared the classification accuracy of our method with that of the standard HOG and Gabor feature extraction methods, as well as several state-of-the-art methods, such as deep neural networks.Our experimental results demonstrate that our proposed method outperforms both HOG and Gabor features in terms of classification accuracy, achieving state-of-the-art performance on all datasets. For example, on the CIFAR-10 dataset, our method achieves a classification accuracy of 89.48%, compared to 87.68% for HOG and 83.25% for Gabor features. On the Caltech-101 dataset, our method achieves a classification accuracy of 92.01%, compared to 87.84% for HOG and 89.50% for Gabor features.C. DiscussionOur proposed hybrid HOG-Gabor features method provides an effective feature extraction framework for object recognition tasks. By combining the complementary information provided by HOG and Gabor features, our method is able to capture both edge andtexture information in an image, resulting in improved classification accuracy. The use of a genetic algorithm to optimize the weights of the feature vectors also enhances the performance of our method.One limitation of our proposed method is its computational complexity due to the use of Gabor filters. Future work could explore ways to reduce the computational cost while maintaining the accuracy of our method. Additionally, our method may not be suitable for real-time recognition tasks due to its high computational requirements, and alternative methods may be needed for such scenarios.Overall, our proposed method provides a novel approach to enhance the performance of feature extraction methods for object recognition, and has potential applications in various fields, such as computer vision, robotics, and autonomous systems.IV. Conclusion and Future WorkIn this paper, we proposed a hybrid HOG-Gabor feature extraction method for object recognition tasks. Our method leverages the strengths of both HOG and Gabor features and overcomes their limitations by combining the information provided by both methods. We used a genetic algorithm to optimize the weights of the feature vectors and achieved state-of-the-art performance on benchmark datasets.Our experimental results demonstrate that our proposed method outperforms both HOG and Gabor features in terms of classification accuracy, indicating the effectiveness of ourapproach. However, our method has some limitations, such as its high computational requirements due to the use of Gabor filters, which could affect its suitability for real-time recognition scenarios.Future work could explore ways to reduce the computational cost of our method while maintaining its accuracy, such as using a smaller number of Gabor filters or reducing the size of the input images. Additionally, we could investigate the possibility of using deep learning methods to further improve the performance of our approach.Another direction for future work could be to apply our method to other object recognition tasks, such as face recognition, object detection, and tracking. Furthermore, our method could also be adapted to other fields, such as medical image analysis, where texture and edge information are essential for accurate diagnosis.In summary, our proposed hybrid HOG-Gabor feature extraction method provides a novel approach to enhance the performance of feature extraction methods for object recognition tasks. We believe that our method has potential applications in various fields and could pave the way for future research in this area.V. AcknowledgementsWe would like to express our gratitude to all the researchers whose works have contributed to the development of feature extraction methods for object recognition. Their dedication and hard work have paved the way for us to conduct our research and make contributions to this field.We would also like to thank our colleagues and mentors for their support and valuable input throughout the development of this paper. Their feedback and insights have been essential in shaping our approach and improving the quality of our work.We would like to acknowledge the support from our institutions for providing the necessary resources and facilities to carry out our research. We also recognize the value of open-source software and datasets which have enabled us to conduct our experiments and validate our approach.Finally, we would like to thank our families and friends for their unwavering support and understanding throughout the duration of our project. Their encouragement has kept us motivated and helped us to persevere through challenging times.In conclusion, we would like to express our deep appreciation to everyone who has contributed to our success in this research endeavor. We hope that our work will make a positive impact in the field of object recognition and inspire future research in this area.。
中英文翻译A configurable method for multi-style license platerecognitionAutomatic license plate recognition (LPR) has been a practical technique in the past decades. Numerous applications, such as automatic toll collection, criminal pursuit and traffic law enforcement , have been benefited from it . Although some novel techniques, for example RFID (radio frequency identification), WSN (wireless sensor network), etc., have been proposed for car ID identification, LPR on image data is still an indispensable technique in current intelligent transportation systems for its convenience and low cost. LPR is generally divided into three steps: license plate detection, character segmentation and character recognition. The detection step roughly classifies LP and non-LP regions, the segmentation step separates the symbols/characters from each other in one LP so that only accurate outline of each image block of characters is left for the recognition, and the recognition step finally converts greylevel image block into characters/symbols by predefined recognition models. Although LPR technique has a long research history, it is still driven forward by various arising demands, the most frequent one of which is the variation of LP styles, for example:(1) Appearance variation caused by the change of image capturingconditions.(2)Style variation from one nation to another.(3)Style variation when the government releases new LP format. Wesummed them up into four factors, namely rotation angle,line number, character type and format, after comprehensive analyses of multi-style LP characteristics on real data. Generally speaking, any change of the above four factors can result in the change of LP style or appearance and then affect the detection, segmentation or recognition algorithms. If one LP has a large rotation angle, the segmentation and recognition algorithms for horizontal LP may not work. If there are more than one character lines in one LP, additional line separation algorithm is needed before a segmentation process. With the variation of character types when we apply the method from one nation to another, the ability to re-define the recognition models is needed. What is more, the change of LP styles requires the method to adjust by itself so that the segmented and recognized character candidates can match best with an LP format.Several methods have been proposed for multi-national LPs or multiformat LPs in the past years while few of them comprehensively address the style adaptation problem in terms of the abovementioned factors. Some of them only claim the ability of processing multinational LPs by redefining the detection and segmentation rules or recognition models.In this paper, we propose a configurable LPR method which is adaptable from one style to another, particularly from one nation to another, by defining the four factors as parameters.1Users can constrain the scope of a parameter and at the same time the method will adjust itself so that the recognition can be faster and more accurate. Similar to existing LPR techniques, we also provide details of detection, segmentation and recognition algorithms. The difference is that we emphasize on the configurable framework for LPR and the extensibility of the proposed method for multistyle LPs instead of the performance of each algorithm.In the past decades, many methods have been proposed for LPR that contains detection, segmentation and recognition algorithms. In the following paragraphs, these algorithms and LPR methods based on them are briefly reviewed.LP detection algorithms can be mainly classified into three classes according to the features used, namely edgebased algorithms, colorbased algorithms and texture-based algorithms. The most commonly used method for LP detection is certainly the combinations of edge detection and mathematical morphology .In these methods, gradient (edges) is first extracted from the image and then a spatial analysis by morphology is applied to connect the edges into LP regions. Another way is counting edges on the image rows to find out regions of dense edges or to describe the dense edges in LP regions by a Hough transformation .Edge analysis is the most straightforward method with low computation complexity and good extensibility. Compared with edgebased algorithms, colorbased algorithms depend more on the application conditions. Since LPs in a nation often have several2predefined colors, researchers have defined color models to segment region of interests as the LP regions .This kind of method can be affected a lot by lighting conditions. To win both high recall and low false positive rates, texture classification has been used for LP detection. In Ref.Kim et al. used an SVM to train texture classifiers to detect image block that contains LP pixels.In Ref. the authors used Gabor filters to extract texture features in multiscales and multiorientations to describe the texture properties of LP regions. In Ref. Zhang used X and Y derivative features,grey-value variance and Adaboost classifier to classify LP and non-LP regions in an image.In Refs. wavelet feature analysis is applied to identify LP regions. Despite the good performance of these methods the computation complexity will limit their usability. In addition, texture-based algorithms may be affected by multi-lingual factors.Multi-line LP segmentation algorithms can also be classified into three classes, namely algorithms based on projection,binarization and global optimization. In the projection algorithms, gradient or color projection on vertical orientation will be calculated at first. The “valleys”on the projection result are regarded as the space between characters and used to segment characters from each other.Segmented regions are further processed by vertical projection to obtain precise bounding boxes of the LP characters. Since simple segmentation methods are easily affected by the rotation of LP, segmenting the skewed LP becomes a key issue to be solved. In the binarization algorithms, global or local methods are often used3to obtain foreground from background and then region connection operation is used to obtain character regions. In the most recent work, local threshold determination and slide window technique are developed to improve the segmentation performance. In the global optimization algorithms, the goal is not to obtain good segmentation result for independent characters but to obtain a compromise of character spatial arrangement and single character recognition result. Hidden Markov chain has been used to formulate the dynamic segmentation of characters in LP. The advantage of the algorithm is that the global optimization will improve the robustness to noise. And the disadvantage is that precise format definition is necessary before a segmentation process.Character and symbol recognition algorithms in LPR can be categorized into learning-based ones and template matching ones. For the former one, artificial neural network (ANN) is the mostly used method since it is proved to be able to obtain very good recognition result given a large training set. An important factor in training an ANN recognition model for LP is to build reasonable network structure with good features. SVM-based method is also adopted in LPR to obtain good recognition performance with even few training samples. Recently, cascade classifier method is also used for LP recognition. Template matching is another widely used algorithm. Generally, researchers need to build template images by hand for the LP characters and symbols. They can assign larger weights for the important points, for example, the corner points, in the4template to emphasize the different characteristics of the characters. Invariance of feature points is also considered in the template matching method to improve the robustness. The disadvantage is that it is difficult to define new template by the users who have no professional knowledge on pattern recognition, which will restrict the application of the algorithm.Based on the abovementioned algorithms, lots of LPR methods have been developed. However, these methods aremainly developed for specific nation or special LP formats. In Ref. the authors focus on recognizing Greek LPs by proposing new segmentation and recognition algorithms. The characters on LPs are alphanumerics with several fixed formats. In Ref. Zhang et al. developed a learning-based method for LP detection and character recognition. Their method is mainly for LPs of Korean styles. In Ref. optical character recognition (OCR) technique are integrated into LPR to develop general LPR method, while the performance of OCR may drop when facing LPs of poor image quality since it is difficult to discriminate real character from candidates without format supervision. This method can only select candidates of best recognition results as LP characters without recovery process. Wang et al. developed a method to recognize LPR with various viewing angles. Skew factor is considered in their method. In Ref. the authors proposed an automatic LPR method which can treat the cases of changes of illumination, vehicle speed, routes and backgrounds, which was realized by developing new detection and segmentation algorithms with robustness to the5illumination and image blurring. The performance of the method is encouraging while the authors do not present the recognition result in multination or multistyle conditions. In Ref. the authors propose an LPR method in multinational environment with character segmentation and format independent recognition. Since no recognition information is used in character segmentation, false segmented characters from background noise may be produced. What is more, the recognition method is not a learning-based method, which will limit its extensibility. In Ref. Mecocci et al. propose a generative recognition method. Generative models (GM) are proposed to produce many synthetic characters whose statistical variability is equivalent (for each class) to that showed by real samples. Thus a suitable statistical description of a large set of characters can be obtained by using only a limited set of images. As a result, the extension ability of character recognition is improved. This method mainly concerns the character recognition extensibility instead of whole LPR method.From the review we can see that LPR method in multistyle LPR with multinational application is not fully considered. Lots of existing LPR methods can work very well in a special application condition while the performance will drop sharply when they are extended from one condition to another, or from several styles to others.多类型车牌识别配置的方法自动车牌识别(LPR)在过去的几十年中的实用技术。
计算技术与自动化Computing Technology and Automation第40卷第1期2 0 2 1年3月Vol. 40,No. 1Mar. 2 02 1文章编号:1003-6199( 2021 )01-0101 — 03DOI : 10. 16339/j. cnki. jsjsyzdh. 202101019基于3次B 样条小波变换的改进自适应阈值边缘检测算法王 煜J 谢 政,朱淳钊,夏建高(湖北工程职业学院建筑与环境艺术学院,湖北黄石435005)摘要:针对含噪声图像边缘提取问题,提出了一种改进NormalShrink 自适应阈值去噪算法。
该算法首先通过小波变换和局部模极大值法提取出可能包含图像边缘特征的小波系数,利用边缘像素之间特殊的空间关系以及噪声在各级小波分解尺度下的不同效应,构建适合各个尺度级的改进NormalShrink 自适应阈值,并依此对提取出的小波系数进行筛选。
实验结果表明,与改进的Candy 算子和传统的NormalShrink 自 适应阈值相比,本方法提取出的图像边缘较为完整清晰,峰值信噪比提升约6 db o关键词:边缘提取;小波变换;自适应阈值;峰值信噪比中图分类号:TP312文献标识码:AAn Improved Adaptive Threshold Edge Detection AlgorithmBased on Cubic B-spline Wavelet TransformWANG Yu f , XIE Zheng,ZHU Chun-zhao ,XIA Jian-gao(School of Architecture and Environmental Art, Hubei Engineering Institute, Huangshi, Hubei 435005, China)Abstract : In order to solve the problem of noisy image edge detection, an improved NormalShrink adaptive waveletthreshold is put forward on the foundation of combining edge detection and denoising . According to the different characteris tics of noise at different wavelet scales and the special spatial relationship between the edge pixels , the algorithm first extract wavelet coefficients which may contain image edge feature by using wavelet transform and local maximum mode, and thenconstruct an improved NormalShrink adaptive threshold of each scale level which is used to select the extracted wavelet coef ficients. Experimental results show that this method can keep imagers edges clear and increase PSNR about 6 db.Key words :edge detection ; wavelet transform ; adaptive threshold ; PSNR图像边缘信息的识别和提取在图像分割、图像 识别等领域有着重要的应用,提取出清晰有效的边缘是一个热点研究方向。
Vol. 29, No. 1Jan. 2021第29卷第1期2021年1月AdvancedTextileTechnologyDOI : 10. 19398/j. a t 202005004引用格式:俞新星,壬勇,支佳雯.织物表面疵点检测方法的设计与实现现代纺织技术,021,9():62 — 67.织物表面疵点检测方法的设计与实现俞新星,任勇,支佳雯(苏州大学应用技术学院,江苏苏州215325)摘要:针对传统织物生产企业中,人工检测织物存在瑕疵检出效率低、误检率高的问题,提出了一种织物表面疵点检测方法。
该方法首先采用高斯滤波、线性归一化以及限制对比度自适应直方图均衡化对织物表面图像进行预处理,从而有效增强图像中的疵点表现细节,然后通过改进的Gabor 优化选择,再对选择后的图像进行初分解,从中挑选出最优滤波图像进行二值化处理,最后运用统计学方法进行疵点判断并获得最终结果。
该方法实现简便、硬件要求低、适应性广,可用于判断织物表面是否含有疵点,并定位疵点。
实验证明,织物表面疵点检测准确率高达95.38%.关键词:织物疵点检测;Gabor 优化选择;直方图均衡化;线性归一化中图分类号:TS103;TP391文献标志码:A 文章编号:1009— 265X(202 1 )01 —0062— 06Design and Implementation of Defect Detection Method for Fabric SurfaceYU Xinxing , REN Yong , ZHI Jiawen(Applied Technology College of Soochow University, Soochow 21 5325 , China)Abstract : To address the problems of low defect, detection efficiency and high false detectionrateof manualfabric detectionin traditionalfabric manufacturing enterprises ,a fabric surface defect, detection method is proposed. For purpose of this method , the Gaussianfilter , linear normalization and limited contrast, adaptive histogram equalization are adopted for preprocessing fabric surface images , to display detect details of the images clearly.Secondly , the selected images are preliminarily decomposed viaimproved optimal Gaborfilter ,with a view to picking outthe ones with the optimalfiltering for binarization processing. Lastly, defect, judgment, is conducted by means of statistical approach , and thefinalresultisobtained.The methodiseasytooperate ,haslow requirementsintermsof hardware , and is of wide adaptability. It can be used to judge the presence of defects onfabricsurface ,andlocatethem.The method is proved to have an accuracy rate of fabric surfacedefectdetectionashighas95.38% throughexperiments.Key words :fabric defect detection ;optimal Gabor filter ; histogram equalization ;linearnormalization收稿日期:2020 —05 —09网络出版日期:2020 —10 —21基金项目:江苏省高校自然基金项目(19KJB520051);江苏高校哲学社会科学研究基金项目(2018SJA2251);江苏省大学生创新创业训练计划项目(201913984009Y)作者简介:俞新星(1998 — )男,江苏如皋人,2017级软件工程专业本科生。
Frangi滤波器和模糊C均值算法相结合的织物瑕疵检测张缓缓;李仁忠;景军锋;李鹏飞;赵娟【摘要】为解决织物瑕疵自动检测问题,提出一种基于Frangi滤波器和模糊C均值算法(FCM)相结合的织物瑕疵检测方法.首先采用均值下采样方法对采集的织物图像进行预处理,以减少织物背景纹理信息对织物瑕疵检测产生的影响;然后通过Frangi滤波器滤波增强织物的瑕疵区域;最后利用FCM处理滤波后的图像,确定织物瑕疵区域的像素和非瑕疵区域像素的聚类中心,并分割出瑕疵区域和非瑕疵区域.结果表明,本文方法检测织物瑕疵种类较多,分割效果较好.与其他方法相比,本文提出的算法利用聚类思想对织物疵点进行分割,无需利用正常织物图像进行阈值计算;另外经过滤波后疵点信息明显增强,使得疵点信息与纹理明显不同,从而使聚类更为准确,增加了检测的准确度.【期刊名称】《纺织学报》【年(卷),期】2015(036)009【总页数】5页(P120-124)【关键词】疵点检测;织物疵点;Frangi滤波器;模糊C均值聚类算法【作者】张缓缓;李仁忠;景军锋;李鹏飞;赵娟【作者单位】西安工程大学电子信息学院,陕西西安710048;西安理工大学自动化与信息工程学院,陕西西安710048;西安工程大学电子信息学院,陕西西安710048;西安工程大学电子信息学院,陕西西安710048;西安工程大学电子信息学院,陕西西安710048;西安工程大学电子信息学院,陕西西安710048【正文语种】中文【中图分类】TP391织物瑕疵的出现很大程度影响着织物质量和纺织企业的经济效益,因此织物的质量检测在纺织行业中占重要地位。
据统计,织物瑕疵对织物销售价格有很大的影响,致使其价格降低45%~65%,严重影响了企业的利润[1]。
目前国内许多企业对织物瑕疵检测的方法落后,检测率低且准确度不高[2-3]。
此外,织物瑕疵种类繁多、形状各异,这更给准确检测织物的瑕疵提出了挑战。
为了适应现代化企业生产的需求,解决传统织物疵点检测方法速度慢,准确率低等缺点, 进一步提高企业的利润,急需开发更先进更智能化的疵点检测方法。