基于图像的变化检测方法研究

发布时间：2018-04-25 01:32

本文选题：变化检测 + Harris角点　；参考：《浙江大学》2017年硕士论文

【摘要】：图像技术在工程应用中使用广泛,随着使用范围的增加,原有的技术手段与相应算法已不能有效解决新遇到的问题,不断发展图像技术并用于实际工程中为人类生活带来各种便利,这是图像时代的发展趋势。论文主要研究基于图像的变化检测方法以及相关技术,并结合实际工程中的相关应用展开论述。论文研究了图像配准算法和灰度校正算法,这两种算法可以减少光照变化等情况带来的干扰。论文详细介绍了几种常用的图像配准算法,分析了这些算法各自的不足。论文主要研究了基于Harris角点的配准算法,着重研究角点的筛选方法,分析了现有的两种筛选方法的不足,并设计了一种基于矩阵特征值比值的筛选方法。论文研究了基于图像的变化检测算法。详细介绍了基于图像灰度差值的变化检测算法和基于LBP直方图互相关系数的变化检测算法并结合实验分析其性能。论文还设计了一种对光照变化不敏感的特征,同时利用该特征对图像进行变化检测。利用该特征对Haar-like特征改进,设计了基于改进的Haar-like特征和随机森林的变化检测方法。最后,论文研究了基于马尔科夫随机场并将其应用于图像的变化检测。论文介绍了基于马尔科夫随机场的图像分割方法,分析了图像分割与图像变化检测问题的相似之处。本文设计了一种基于马尔科夫随机场的融合方法,该方法可以将多种变化检测算法的检测结果图像融合,融合后的最终结果比单一的检测结果更好。论文结合上述研究内容设计了一个基于图像的变化检测系统,并给出实验结果。
[Abstract]:Image technology is widely used in engineering applications. With the increase of the scope of application, the original technical means and corresponding algorithms can not effectively solve the new problems encountered. It is the development trend of image age to develop image technology and bring all kinds of convenience for human life in practical engineering. This paper mainly studies the image-based change detection methods and related technologies, and discusses the related applications in practical engineering. In this paper, image registration algorithm and gray level correction algorithm are studied, which can reduce the interference caused by illumination change. Several common image registration algorithms are introduced in detail, and their shortcomings are analyzed. This paper mainly studies the registration algorithm based on Harris corner, focuses on the corner selection method, analyzes the shortcomings of two existing screening methods, and designs a selection method based on matrix eigenvalue ratio. The change detection algorithm based on image is studied in this paper. The change detection algorithm based on image gray difference and the change detection algorithm based on the correlation number of LBP histogram are introduced in detail. The performance of the algorithm is analyzed in combination with experiments. A feature which is insensitive to light change is designed and used to detect the change of image. Using this feature to improve the Haar-like feature, an improved Haar-like feature and a change detection method based on random forest are designed. Finally, based on Markov random field, we apply it to image change detection. This paper introduces the method of image segmentation based on Markov random field and analyzes the similarity between image segmentation and image change detection. In this paper, a new fusion method based on Markov random field is designed. This method can fuse the image of various change detection algorithms. The final result of fusion is better than that of single detection. In this paper, a change detection system based on image is designed, and the experimental results are given.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】