Image stitching is an important branch of image processing, and image stitching technology is widely used in academia and industry. As the previous step of stitching, the accuracy of image registration plays a key role in the process of stitching. Common algorithms such as SIFT and Harris are often used for feature extracting and feature matching, but the time complexity is high, at the same time, the extracted feature points are mostly pixel level, the accuracy of which is low. This paper proposes an improved Harris detection algorithm, by introducing bilinear interpolation algorithm, the feature points matching is located to the sub-pixel level; In the image stitching stage, iterative optimization is used to refine perspective transformation matrix. The experimental results show that the sub-pixel registration and stitching technology based on interpolation and iterative optimization greatly reduces the time complexity of feature extraction and effectively improves the accuracy of stitching. Key Words: Image registration, Image stitching, Sub-pixel, Interpolation algorithm, Iterative optimization