Wide Baseline Image Stitching with Structure-Preserving

This paper presents a novel stitching approach for wide-baseline images with low texture. Firstly, a three-phase feature matching model is applied to extract rich and reliable feature matching, in the case of low texture, our line matching and contour matching will compensate for the poor quality of point matching. Then, a structure-preserving warping is performed, by defining several constraints and minimizing the objective function to solve the optimal mesh, with which we obtain multiple affine matrices to warp images. Furthermore, we synthetically consider alignment error, color difference and saliency difference to find the optimal seam for image blending. Experiments both on common data sets and challenging surveillance scenes illustrate the effectiveness of the proposed method, and our approach has outstanding performance when compared with other state-of-the-art methods.