In order to solve the problem of low efficiency and poor robustness of traditional image mosaic algorithm, an improved SIFT-based feature matching algorithm for image stitching were put forward. Firstly enhanced the images by using the discrete cosine transform，quantization, inverse discrete cosine transform and inverse quantization in the image preprocessing stage, and then proposed the pre-inspection strategy in view of the RANSAC algorithm iteration increasing when computing the transformation models. The experimental results showed that the improved algorithm obtained the quite ideal effect without changing the matching precision.