The preprocessing to panoramic stitching mainly includes extracting features form sample images and matching. Then to images of overlapping. In this approach,an extended Harris corner detector is proposed by us to accurately extract feature points from an image and to assign a feature descriptor for each of them. After that,we suggests a new feature indexing algorithm based on wavelet coefficients which enables comparing features in neighboring images more efficiently. Experiments show that feature matches are accurate and efficient,and seamless panoramic image stitching can be produced.