In order to solve the problem of large computation and long time consuming of traditional image mosaic algorithm, a novel SURF image mosaic method based on wavelet transform is proposed in this paper. Firstly, adopting the Haar wavelet image preprocessing method to get the second order decomposition, the low frequency components of image are obtained. And by using wavelet gradient vector, the feature point can be extracted for low frequency image overlap region. So, the transformation parameters of characteristic point are quickly acquainted in low frequency images, which can guide to select the feature point extraction in high frequency images. Based on that, an improved SURF image matching algorithm is proposed by using the properties of the single direction matching and the orientation coherence, which can effectively eliminate the mismatched point pair to improve the accuracy and realtime performance of feature point matching. Finally, two experiments are used to verify the feasibility and effectiveness of the proposed results.