关键词: 加速鲁棒特征（SURF）算法, 图像拼接, 小尺寸图像, 参数配置, 特征点匹配, 窗口滤波器, 特征点子区域

Abstract: In the field of image stitching, SURF algorithm is widely used for its excellent real-time procedure and robust. Due to the fixed parameter settings in interest point detection and description, the adjustability of SURF algorithm to different applying situation is relatively weak. In order to improve the range of SURF application, this paper looks into three aspects in the algorithm: the weight of box filter, the sub-region selection and the Haar wavelet sampling point range within the sub-region. In comparing the real-time quality, the stitching accuracy and the robust quality of the algorithm, the experiments using variable control method propose a parameter setting strategy based on small scale image which is widely used in nowadays stream media. In analyzing the interest points match rate and efficiency, the strategies proposed in characterizing the algorithm can extend the use of SURF in real-time field.

Key words: Speeded Up Robust Features（SURF） algorithm, image stitching, small scale image, parameter settings, interest points matching, box filter, sub-region of interest points