SIA OpenIR  > 光电信息技术研究室
图像质量表征与目标识别处理技术研究
其他题名A Research on Image Quality Characterization and Target Recognition Processing
肖传民1,2
导师史泽林
分类号TP391.41
关键词图像质量表征 梯度分布 图像匹配 边缘检测 目标识别
索取号TP391.41/X47/2015
页数103页
学位专业模式识别与智能系统
学位名称博士
2015-05-28
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门光电信息技术研究室
摘要自动目标识别(ATR)已成为现阶段和未来武器系统的重要组成部分,ATR技术研究是决定下一代自主式灵巧武器能否取得成功的关键。图像质量表征和目标识别处理技术研究是ATR系统工程化的研究难点和热点,是决定ATR系统能否实现和很好发挥的基础和核心技术。图像质量表征可为ATR的开发和评价提供参考,与ATR算法直接相关,影响所有可能的设计方案;目标识别处理技术包括图像预处理、特征选择与提取、目标匹配等关键技术,是ATR算法的核心。因此,本文围绕图像质量表征与目标识别处理技术等问题展开了深入研究。第1章介绍研究背景与意义,分析了自动目标识别、图像质量表征、目标匹配识别技术的研究现状,归纳总结了论文的研究内容与研究难点。第2章引入梯度分布特征的图像质量表征。首先分析了杂波的定义、杂波在目标获取性能模型中的引入和现有杂波度量尺度;然后提出了一种引入梯度分布特征的图像背景杂波度量尺度,该尺度用梯度方向直方图表征目标结构特征,对结构相似性图像杂波度量进行加权。实验表明该杂波度量尺度在相关系数、均方误差等方面优于其他现有杂波度量方法,以此为基础的目标获取性能目标探测概率、虚警率、目标搜索时间方面与观察者的一致性较高,具有较高的预测精度。第3章围绕目标结构特征的保持和图像匹配的效率问题展开研究,提出一种基于张量子空间降维的边缘图像匹配算法。该算法针对传统基于向量子空间降维的图像匹配算法易丢失像素间邻域关系和计算量大的问题,通过双边投影变换提取边缘图像的张量子空间,在降低特征空间维数的同时保持边缘特征之间的邻域关系,同时采用边缘膨胀后的互相关度量模板与实时图的相似性。标准人脸数据库和红外实时图像的匹配实验结果表明,该算法在匹配时间、匹配正确率、匹配精度三方面较传统基于向量子空间的匹配算法有显著的性能提高,并且该算法具有较高的工程应用价值。第4章和第5章围绕边缘特征提取问题,研究了图像增强预处理和稳定边缘检测算法。第4章研究了图像增强方法。重点研究了能够改善图像对比度、锐化图像边缘的Retinex图像增强方法,提出了一种基于小波变换的Retinex图像增强算法。首先对Retinex算法的光晕伪影问题进行了机理分析,采用双边滤波代替高斯滤波进行照度图像估计,解决了光晕伪影问题;针对双边滤波的参数选取难题,采用图像小波变换分解估计噪声方差,进而自适应确定双边滤波器的参数。通过客观指标评价和不同增强算法的比较实验,证明该算法增强了图像边缘,改善了图像质量。第5章面向目标识别处理技术中的特征提取问题,开展了边缘检测算法研究。首先详细研究了边缘的物理意义、检测原理、检测准则和最优边缘检测算子数学模型。然后为了抑制背景和纹理等不稳定边缘,突出目标边缘信息,为自动目标识别提供质量较好的边缘图像,提出了稳定边缘检测算法。该算法以边缘对比度、边缘密度和边缘尺度3个独立的基本边缘信息为约束条件,有效地抑制了背景和纹理虚假边缘,突出了目标边缘信息。首先采用稳定边缘检测算法对参考图像和待匹配图像进行边缘检测,然后在边缘图的基础上进行相位相关匹配,提出了一种基于边缘相位相关的图像匹配算法。实验结果证明该算法提高了匹配精度。
其他摘要Automatic Target Recognition (ATR) has become a significant part in weapon system both nowadays and in the future. The research on ATR is a crucial part which determines the success of autonomous smart weapons. The characterizations of image quality and target recognition technology research are hot and difficult topics in the research of ATR system, and they are also the foundation and kernel that determines whether the ATR system could be achieved and function well or not. The characterization of image quality highly related with the ATR algorithm provides reference for the development and evaluation of the ATR, and also affects all possible design schemes. The technology of target recognition processing includes the key technologies that contain image pre-processing, feature selection and extraction, and target matching. Moreover, it’s the core of the ATR algorithm. Therefore, this paper researches in-depth on issues such as the characterizations of image quality and target recognition technology.The chapter 1 introduces the researching background and significance, and analyzes researching status of the automatic target recognition, the characterization of image quality, and the target matching recognition.The chapter 2 presents the gradient distribution characterization of image quality. Firstly, it makes a full analysis on the basis of biology vision, the methods of characterization, and the existing clutter measurement method. Secondly, it introduces an image background clutter metric scale based on the gradient distribution, and this scale makes the histogram of oriented gradient as its structural feature and weighs the Image Structural Similarity clutter metric. Experimental results show that the proposed method exceeds other current clutter measurement method in both correlation coefficients and mean-square errors.The chapter 3 researches on target recognition application related to machine vision. Considering the efficiency problem of edge image matching recognition, the paper puts forward edge image matching algorithm based on tensor subspace dimensionality reduction. Aiming at the questions of easily losing relationships between pixels and intensively computational problems using traditional vector subspace methods, the algorithm extracts Tensor subspace by employing two-sided projection transformation in edge images, reducing dimension of feature space and preserving the relationships between edge pixels, and the algorithm measures the similarity between template and real-time image by calculating the correlation of dilated binary images. Experimental results on the standard face database and real IR images show that the new algorithm can improve the computational efficiency remarkably,and has a higher matching rate and matching precision than traditional vector subspace methods. The proposed algorithm can also be applied in cluttering and partially occluded circumstances. After the robustness analysis and application prospects, it shows that the algorithm is rather valuable in engineering application.The chapter 4 researches the image edge enhancement methods that highlight the edge features. It focuses on the research of Retinex image enhancement method improving the image contrast and sharpening the image edge, and puts forward the Retinex image edge enhancement algorithm base on wavelet transform. Through analyzing the problem of halo artifacts to Retinex algorithm, instead of Gaussian filter, using bilateral filtering for illumination image estimation, this part solves the problem of halo artifacts; For the parameter selection problem of bilateral filtering, wavelet transform is used to decompose and estimate the noise variance of image, and it determines the parameter of bilateral filter adaptively. Designing the experiment of analyzing subjective visual effect, evaluating objective indicator and comparing different enhancement algorithms, it shows that the algorithm improves the information of gradient distribution, highlights the image edge features and improves image quality.The chapter 5 researches the edge detection algorithm for the characterization extraction problems related to target recognition processing. First of all, this part studies particularly on the physical meaning of edge, detection theory, detection principle, and the mathematical model of optimal edge detection operator. Then aiming at the problem of the positioning accuracy of the edge detection process, the paper puts forward the edge detection algorithm grounded on visual saliency, which effectively suppresses the false edge of background and texture, highlights the target edge information, positions accurately using edge detection. Finally, the edge image of visual saliency is applied to the image matching algorithm of phase congruency based edge, and it improves the matching precision.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/16779
专题光电信息技术研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
肖传民. 图像质量表征与目标识别处理技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2015.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
图像质量表征与目标识别处理技术研究.pd(5391KB) 开放获取CC BY请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[肖传民]的文章
百度学术
百度学术中相似的文章
[肖传民]的文章
必应学术
必应学术中相似的文章
[肖传民]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。