【正文】 2 國(guó)內(nèi)外 研究現(xiàn)狀 使用一組局部興趣點(diǎn)來進(jìn)行圖像匹配的發(fā)展可以追溯到 1981年 Moravec在立體匹配中使用的角探測(cè)器?；?SIFT算法的圖像匹配技術(shù)在圖像處理技術(shù)中起著十分關(guān)鍵的作用，它是圖像處理技術(shù)得以發(fā)展的一個(gè)重要基礎(chǔ)。 圖像匹配的核心問題是將同一目標(biāo)在不同時(shí)間、不同分辨率、不同光照、不同位姿情況下所成的像相對(duì)應(yīng)。 特征的提取是圖像匹配技術(shù)的最重要組成部分，一幅圖像中可以檢測(cè)到多個(gè)關(guān)鍵點(diǎn)，我們需要消除一些不需要的點(diǎn)，保留特征性較強(qiáng)的點(diǎn)，用這些點(diǎn)來進(jìn)行圖像匹配。提取 特征點(diǎn)的算子叫做興趣算 子或有利算子，即從圖像中提取出一些能代表圖像信息的特征點(diǎn)，運(yùn)用這些特征點(diǎn)來對(duì)圖像進(jìn)行處理，以便達(dá)到目的 。點(diǎn)特征可以應(yīng)用于諸如圖像的配準(zhǔn)與匹配 ,目標(biāo)描述與識(shí)別 , 光束計(jì)算 ,運(yùn)動(dòng)目標(biāo)跟蹤、識(shí)別和立體像對(duì) 3D建模等眾多領(lǐng)域 [2]。 基于 SIFT算法的圖像匹配技術(shù)是 1999年 British Columbia大學(xué)大衛(wèi) .勞伊教授總結(jié)了現(xiàn)有的基于不變量技術(shù)的特征檢測(cè)方法，并正式提出了一種基于尺度空間的圖像局部特征描述算子－ SIFT， 并在 2020年被加以完善。正因?yàn)槠鋺?yīng)用的廣泛性，新的應(yīng)用和新的要求逐步產(chǎn)生使得匹配算法的研究逐步走向深入 [1]。 關(guān)鍵字： 圖像匹配 ，尺度空間，穩(wěn)定性， SIFT 算法，物體識(shí)別 基于 sift 算法的圖像匹配方法 II ABSTRACT Image matching refers to the process of identify homonymy points between two or more images by a certain algorithm. Generally speaking, Image matching technology is of important, not only in the application of image processing technology, but also in the field of pattern information. Besides, it is also the basis of some other technology of image analysis. Therefore, it would seem that further investigation about image matching technology is needed. Image matching is a multistep process. Different matching algorithm would have different calculate steps and results. In addition, SIFT is an algorithm based on scalespace, which feature is the local features of the image, its rotation, scale zooming, brightness variation have the advantage of invariability, and the c hange of viewpoint ,pattern transformation, noise also has good trait of stability. Meanwhile, the SIFT algorithm has the characteristics of good uniqueness, huge information, faster running speed and so on. This article presents that the method of image matching what based on SIFT algorithm, and summarize that the implementation procedure of SIFT algorithm as well as explores the matching results of SIFT in the aspect of brightness variation, scale variation and Rotation , it also discusses the way of object recognition by use of SIFT algorithm. Keywords: image matching, scalespace, stability, SIFT algorithms, object recognition 基于 sift 算法的圖像匹配方法 III 目 錄 1 緒論 .......................................................................................................................... 1 研究的背景及意義 .......................................................................................... 1 國(guó)內(nèi)外研究現(xiàn)狀 .............................................................................................. 2 研究?jī)?nèi)容和目標(biāo) .............................................................................................. 3 章節(jié)安排 ......................................................................................................... 3 2 圖像匹配 ................................................................................................................ 5 圖像匹配技術(shù)概述 ......................................................................................... 5 圖像匹配技術(shù)的定義 ..................................................................................... 5 圖像匹配算法分類 ......................................................................................... 6 本章小結(jié) ......................................................................................................... 6 3 尺度空間理論 ....................................................................................................... 7 尺度空間理論 ................................................................................................. 7 高斯尺度空間 ................................................................................................. 7 本章小結(jié) ......................................................................................................... 8 4 SIFT 特征匹配 ...................................................................................................... 9 圖像的初始 ................................................................................................... 10 尺度空間極值檢測(cè) ....................................................................................... 10 精確定位特征點(diǎn)的位置和所在的尺度 ....................................................... 14 確定特征點(diǎn)的主方向 ................................................................................... 17 生成 SIFT 特征向量 ..................................................................................... 18 SIFT 特征向量的匹配 .................................................................................. 20 本章小結(jié) ....................................................................................................... 21 5 實(shí)驗(yàn)結(jié)果及分析 ................................................................................................ 23 SIFT 特征提取實(shí)驗(yàn)結(jié)果 .............................................................................. 23 特征匹配實(shí)驗(yàn)結(jié)果 ....................................................................................... 24 圖像匹配應(yīng)用 —— 物體識(shí)別 ....................................................................... 29 本章小結(jié) ....................................................................................................... 36 6 總結(jié)與展望 .......................................................................................................... 38 總結(jié) ............................................................................................................... 38 展望 ............................................................................................................... 38 參考文獻(xiàn) ................................................................................................................... 40 致謝 ............................................................................................................................ 41 附錄 ............................................................................................................................ 42 基于 sift 算法的圖像匹配方法