基于局部不变特征的目标识别方法研究
文献类型：学位论文
| 作者 | 姜永兵 |
| 学位类别 | 硕士 |
| 答辩日期 | 2011-05-31 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 北京 |
| 导师 | 彭启民 |
| 关键词 | 局部不变特征，视觉基元，视觉词汇，目标模型，半邻域特征组合，模式总结，模式分解，稳定模式，目标识别 |
| 学位专业 | 信息系统集成 |
| 中文摘要 | 目标识别是计算机视觉研究的核心问题之一。基于局部不变特征构建的识别算法可以显著提高视觉应用系统的性能，已经被广泛应用于图像目标识别、图像检索等领域。 本文首先对局部不变特征检测子和描述子的原理进行研究,分析比较不同检测子检测到特征的结构、精确度、不变性、重复性等特性，确定各种检测子的适用场景及选择策略，并对常用局部不变特征描述子的提取方法和适用场景进行分析，尤其对视觉词汇的不足及解决方案进行了阐释和总结。 然后，对局部不变特征在目标建模中的应用方式进行分析比较，提出适用性更强的目标建模方法。 接着，对所提目标建模方法中的局部不变特征组合提取和特征组合优化等问题进行了研究，在特征组合提取方法中提出了对尺度、旋转、仿射等变换鲁棒的局部不变特征组合提取方法，在特征组合优化中通过对数据挖掘频繁项集的分解和总结算法的研究，设计出适合于视觉数据中特征组合优化的模式分解和总结算法。 最后，提出了稳定模式的目标模型，该模型包含两种具有较强表征力和区分力的图像中层表示模式：类间共用稳定模式(Inter-Class Common Stable Pattern)和类内特殊稳定模式(Intra-Class Special Stable Pattern)，通过对由图像半邻域特征组合挖掘得到的频繁项集进行模式分解、统计学过滤、模式总结及模式组成项间几何关系的建模得到。将稳定模式引入到目标识别框架，实验得到了相比同类方法更好的结果。 |
| 英文摘要 | Object recognition is one of the core problems in computer vision. Recognition algorithms based on local invariant features have been widely used in object recognition, image retrieval areas for its high efficiency in computer vision application systems. This paper focuses on explaining and recognizing objects based on local invariant features. The highlights and main contribution of the dissertation are as follows. Firstly, the principles of local invariant feature detectors and descriptors are deeply studied. Detectors’ application areas and selection criterions are proposed by analyzing and comparing their characteristics such as local structure, accuracy, invariance, repeatability and so on. Analyze the extraction methods and application areas for some frequently used descriptors; especially give a detailed explanations and summarizations for the deficiencies and solutions of visual word/texton. Secondly, a more robust object model is proposed by comparing and analyzing object models based on local invariant features. Then, the problems of extract and optimize local invariant feature compositions are investigated. A feature composition extraction method robust to scale, rotate, affine transformations is proposed, and optimize algorithms suitable for visual data are proposed based on the study of pattern decomposition and pattern summarization in data mining. Last, an object model named Stable Pattern is proposed, this pattern includes two image mediate-level representations: Inter-CSP (Inter-Class Common Stable Pattern) and Intra-SSP (Intra-Class Special Stable Pattern). The details of processing is given which can be divided into pattern decomposition, statistic-filtering, pattern summarization and item-based geometric relation modeling on frequent itemsets mined from image semi-local features. And when the stable patterns are introduced into recognition framework, the experiment results are prior to classical methods. |
| 学科主题 | 计算机应用 ; 计算机应用其他学科 |
| 语种 | 中文 |
| 公开日期 | 2011-06-11 |
| 源URL | [http://124.16.136.157/handle/311060/10440]  |
| 专题 | 软件研究所_综合信息系统技术国家级重点实验室 _学位论文 |
| 推荐引用方式 GB/T 7714 | 姜永兵. 基于局部不变特征的目标识别方法研究[D]. 北京. 中国科学院研究生院. 2011. |
入库方式： OAI收割
来源：软件研究所
浏览0
下载0
收藏0
其他版本
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。