图像匹配技术在计算机科学领域中占有非常重要的地位，目前已经被广泛的应用在社会生活生产的各个领域。本文致力于图像匹配技术的研究，首先对图像匹配技术的基本概念、研究背景和意义以及发展现状作了阐述，然后针对SURF图像匹配算法存在仿射变换敏感的问题，提出一种基于MSER和SURF的图像特征匹配算法，并且给出了实验证明。接着本文针对传统的图像匹配算法在实际的应用中普遍存在搜索范围广、检测到的无关特征点多的问题，提出一种基于显著性区域检测的特征匹配算法。最后成功地将提出的两种算法移植到嵌入式平台上。本文的主要研究工作与贡献包括：

（1）对MSER区域检测算法以及SURF算法进行了深入的研究。本文首先分析得出SURF算法是一种对仿射变换敏感的图像匹配算法，然后利用MSER区域检测算子具有完全仿射不变性的特点。然后本文巧妙的将MSER区域特征与SURF特征点结合，设计了一种能将MSER区域转换为方便SURF描述的点的方法，并且使用SURF描述子对所有的融合特征点进行描述，并且最后通过实验对比发现该新算法较原算法具有较大优势。

（2）深入地研究了基于频率调谐和基于最大对称包围的显著性检测算法，为了尽可能的使匹配点对落在显著性目标上，本文结合显著性检测算法和SURF算法提出一种基于视觉显著性检测的特征匹配算法。该算法首先利用视觉显著性算法对图像进行预处理获得显著图，然后设计了一套分割显著性区域的方法，令图像匹配算法的搜索范围限定在显著性区域内，增强了图像匹配的针对性和准确性。

（3）本文还搭建了基于嵌入式linux的实验平台，并且利用该平台完成了本文提出的两种算法的实验，并给出实验图像。

In the field of computer science, image matching is a very important technology, and it has been widely used in all areas of social life production. This paper is devoted to the research of image matching. Firstly, the basic concept, background and significance of the image matching has been introduced briefly, and this paper focused on the problem of poor affine invariance of SURF (speeded up robust features)  algorithm. Then we aimed at this problem, presented a new algorithm which is based on MSER and SURF (It named MSER-SURF in this paper). Then the experiment proved had been gave for the new algorithm. And we also found that many classical image matching algorithms have the same problems that the searching range is very large and irrelevant feature points is great. So we proposed a new feature matching algorithm which is based on salient region detection. And finally, we implement those two algorithms in the embedded platform successfully. The main research contents and contributions of this paper are:

(1) The MSER region detection algorithm and SURF algorithm has been studied deeply. We deeply analysis features of SURF algorithm, found that the SURF algorithm is efficient but with poor affine invariance. And the MSER region detection algorithm is robust for affine invariance. So we combine the MSER region features with the SURF feature points cleverly. A method to transform MSER region to the point which is convenient to describe for SURF descriptor had been proposed. And then all the syncretic features had been described by using SURF descriptor. Finally, we found that the new algorithm has great advantages over the original algorithm through experiments.

(2) The frequency-tuned salient region detection algorithm and saliency detection algorithm using maximum symmetric surround had been analyzed profoundly. In order to make the matching pairs locate on the salient target as far as possible, we combined the advantages of salient region detection algorithm with SURF algorithm, and proposed a new method. The new method which is based on salient region detection algorithm preprocesse the image to obtain the salient image firstly. And we designed a method to get the regions of saliency. Lastly, we implement SURF algorithm in the regions of saliency, so the regions of saliency are the active area of the image matching algorithm. This new method can improve the specificity and accuracy.

(3) An experimental platform based on embedded linux has been established in this paper. Then we implement our new methods in this platform, and gave experimental results.