Abstract: A new algorithm combines with Harris angular points, SURF algorithm and K-Means algorithm for accurately matching and stitching microscope images which are characterized by a great number of local feature points. It is an improved algorithm based on traditional algorithm. Specific procedures are as follows. Firstly, feature points in a microscope image are extracted by Harris algorithm and SURF descriptors are constructed to make rough matching through the nearest neighbor algorithm. Next, K-Means clusters rough-matched feature points into groups and obtains clustering center. Therefore, feature points in effective clustering groups are got. The last step is to make precise matching to effective feature points. On this basis, slope consistency and distance consistency of matched feature point pairs are verified so that precise feature point pairs are obtained. As the result, the new algorithm is with high robustness and stability, which can decrease image joint errors and shorten image joint time in image matching and stitching with a great number of local feature points. It is convinced that this new algorithm applies to real-time microscope image matching and stitching well.

Key words: image stitching, Scale-Invariant Feature Transform（SIFT）, Speeded Up Robust Features（SURF）, Harris, K-Means, consistency check

摘要： 针对传统图像拼接算法不适用于局部特征点多的微观图像实时拼接问题，结合Harris角点、SURF算法和K-Means算法提出了一种改进的算法。具体的算法流程如下：通过Harris角点提取微观图像中的特征点，并在形成SURF描述子后利用最近近邻算法对这些特征点进行粗配准。通过K-Means算法对初次配准的特征点进行聚类分簇获取聚类中心，并提取有效聚类区域的特征点。对有效的特征点进行精确配准，并校验配准后特征点的斜率一致性和距离一致性，从而实现精确的特征点匹配。实验结果证明，该算法克服了特征点多造成图像拼接时间长和拼接误差大的问题，具有较强的鲁棒性和稳定性，可应用于微观图像实时拼接领域。

关键词: 图像拼接, 尺度不变特征变换（SIFT）, 加速稳健特征（SURF）, Harris, K-Means, 一致性校验