Seamless image stitching using structure deformation with HoG matching

In this paper, we propose an efficient image stitching using structure deformation. We use image stitching based on common stitching algorithms such as speeded up robust features (SURF) feature detection, approximated nearest neighbor (ANN) matching and random sample consensus (RANSAC) parameter estimation. And we use homography similarity to identify if input images have enough correlation. To reduce structure misalignment, we use double-seam selection. Through local maximum in gradient domain, we find 1-D feature points along with each seam. To find matching points, we used histogram of oriented gradients (HoG) which find matching points robustly. In addition, we use multi-band blending algorithm to remove intensity difference efficiently. In the experimental results, we compare with common stitching algorithms and existing structure deformation algorithms.