An SIFT-Based Fast Image Alignment Algorithm for High-Resolution Image

To solve the problem of high-resolution image alignment time overhead, an SIFT-based fast algorithm is presented. The overlap region images computed in detail by phase correlation to avoid a lot useless calculations non-overlapping region. After distribution feature points determined difference Gaussian through formula derivation, total number limited. more stable spatially distributed for obtained due expanded detection range extreme and added non-maximum suppression. It noteworthy that descriptor calculated method down-sampling. And circular constructed with only 56-dimensional point generation stage, which makes matching shorter. This indicates calculation faster lower dimensions new algorithm. In addition, experimental results show average (9.60s, 13.46s, 15.81s) proposed 0.86%, 0.43%, 0.10% SIFT algorithm, respectively. overall speed 2–3 orders magnitude than can overhead. provides good stitching quality shows excellent performance compared several existing algorithms at current. has potential application value real-time stitching.