基于局部特征的目标自动识别

	基于局部特征的目标自动识别
	王飞宇
学位类型	硕士
导师	贾平
	2014-07
学位授予单位	中国科学院大学
学位专业	机械电子工程
摘要	图像背景复杂、目标占像素数量少导致目标难以描述，是远距离目标识别的难点。本文针对远距离目标提出了一种基于局部特征的识别算法，主要步骤包括局部特征提取、局部特征编码和特征分类。本文首先比较了几种经典的关键点检测算法，表明带有高斯滤波的检测器会造成的图像细节损失、不利于远距离目标提取，进而提出利用尺度空间AGAST角点检测器进行关键点检测；同时，本文通过数学推导改进了SURF描绘器，提高了其抗旋转性。在此基础上，本位提出利用尺度空间AGAST角点检测器和改进的SURF描绘器提取局部特征。然后，本文算法利用基于空间金字塔匹配(SPM)的特征袋编码(BoF)对局部特征进行编码；得到的合成特征能够更加有效地描述目标。最后，本文算法利用径向基函数(RBF)核的支持向量机分类器进行特征分类，使分类时间达到毫秒级别。实验结果表明，本文提出的识别算法的计算时间(163.9ms)低于基于SURF特征的算法(213.4ms)；对视角变化目标(96.61%)和光照变化目标(96.88%)的识别率高于基于SIFT的算法，而与基于SURF的算法的接近；对尺度变化目标的识别率明显地高于基于SIFT或SURF的识别算法，对于7系数降采样获得的测试图集的识别率超过50%。本文对实验结果进行了分析，并总结本文算法适用于远距离目标的识别。
其他摘要	The difficulty of automatic long-distance target recognition lies in the background cluttered and relatively less pixel the targets have, which factors make long-distance targets difficult to describe. This dissertation proposed a recognition algorithm using local features against long-distance targets, its procedure includes local feature extraction, local feature encoding and feature classification. First, several of the typical key-point detectors are compared, and the statement is made that detectors with Gaussian blur are apt to filter out some of the necessary details and are thus disadvantageous to long-distance target description. Scale-space AGAST corner detector is therefore used. A mathematical derivation is also made to improve the SURF descriptor, and the improvement has gained the descriptor a better resistance to image rotation. Based on these studies, the proposed algorithm therefore extracts local features using scale-space AGAST detector and the improved SURF descriptor. The local features are then encoded using the Bags of Features algorithm based on Spatial Pyramid Matching. The encoded features characterize the targets more effectively. Finally, Support Vector Machine with Radial Basis Function kernel is used for feature classification, allowing for a millisecond-level performance. The experimental results have shown that the proposed algorithm has reached higher computation efficiency (163.9ms) than its counterpart using the original SURF feature (213.4ms); a relatively high recognition rate against targets with view-point change (96.61%) or illumination change (96.88%); and a recognition rate of above 50% against targets with scale change which are obtained via image down-sampling by a factor of 7, far exceeding that of the algorithms using either SIFT or SURF. An analysis over the results is made, and the conclusion can be drawn that the proposed algorithm is well applicable to long-distance target recognition.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/41477
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	王飞宇. 基于局部特征的目标自动识别[D]. 中国科学院大学,2014.

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
王飞宇.pdf（3050KB）			开放获取	CC BY-NC-SA	请求全文