基于机器学习的图像匹配方法研究

发布时间：2018-04-24 01:04

本文选题：机器学习 + 图像匹配　；参考：《沈阳理工大学》2016年硕士论文

【摘要】：随着智能化制造的大力发展,机器视觉已经成为多种领域不可或缺的技术之一,而借助计算机对图像分析和处理的能力模拟人类视觉引导和控制机械结构动作,必然会成为智能制造的趋势。图像匹配技术作为机器视觉中的一项分支,是多种图像处理技术的关键步骤,匹配程度的好坏通常直接决定系统整体性能。但现有匹配技术已无法满足日益出现的新需求,急需新的理论。因此,本文针对图像匹配技术的研究进行了如下工作:1)首先对现有图像匹配技术进行综述,分析了图像匹配技术研究的目的和意义,按照基于像素灰度的图像匹配方法和基于特征的图像匹配方法分别总结当前国内外的研究现状,为后续研究工作奠定理论基础。2)针对基于像素灰度的匹配技术展开研究,在介绍了常见匹配算法和算子的基础上,着重介绍一种被多种领域广泛应用的技术“Lucas-Kanade算法”,本文将对其进行深入学习和推导。3)针对基于特征的图像匹配技术展开研究,在介绍了经典的几类特征提取算子的基础上,着重分析一种基于图像边缘特征的匹配算法,经测试其对复杂场景、光照变化、遮挡等均具有鲁棒性。4)针对点模式匹配问题展开研究,对现有点模式匹配方法进行了详细分析,针对点模式匹配算法中存在的缺陷并结合多种机器学习方法提出了一种基于遗传算法的非刚性点集匹配算法,并通过对比实验验证其有效性。本文着重介绍的两种图像匹配技术能为一线工程人员提供直接有效的参考;提出的基于遗传算法的点集匹配方法不仅展现了优秀的性能,更扩展了遗传算法的应用领域,为智能学习方法与传统匹配算法相结合的道路提供了一点参考。
[Abstract]:With the development of intelligent manufacturing, machine vision has become one of the indispensable technologies in many fields. It will certainly become the trend of intelligent manufacturing. As a branch of machine vision, image matching technology is a key step of many image processing techniques. The degree of matching usually directly determines the overall performance of the system. However, the existing matching technology can not meet the new demand and needs new theory. Therefore, in this paper, the research of image matching technology is as follows: 1) first of all, the existing image matching technology is reviewed, and the purpose and significance of image matching technology research are analyzed. According to the image matching method based on pixel gray level and the image matching method based on feature, this paper summarizes the current research status at home and abroad, establishes the theoretical foundation for further research work. Based on the introduction of common matching algorithms and operators, this paper focuses on the introduction of Lucas-Kanade algorithm, which is widely used in many fields. Based on the introduction of several classical feature extraction operators, a matching algorithm based on image edge features is analyzed. In view of the point pattern matching problem, the existing point pattern matching methods are analyzed in detail. In this paper, a non-rigid point set matching algorithm based on genetic algorithm is proposed to overcome the shortcomings of point pattern matching algorithm and combined with various machine learning methods. The effectiveness of the algorithm is verified by a comparative experiment. The two image matching techniques introduced in this paper can provide direct and effective reference for front-line engineers. The point set matching method based on genetic algorithm not only shows excellent performance, but also extends the application field of genetic algorithm. It provides a reference for the combination of intelligent learning method and traditional matching algorithm.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP181

【相似文献】