Algorithm 1 Stitching two images.

Input: a target image $I_a$ and a reference image $I_b$. Output: a stitched image.   1:Match point and line features between $I_a$ and $I_b$ to obtain $\{p_a,p_b\}$ and $\{l_a,l_b\}$ via feature correspondences increase.   2:Calculate a global homography $H_{gh}$ and a global similarity matrix $H_{gs}$ via dual-feature.   3:Calculate a hybrid warp $H_{init}$ by Equation (18) and Equation (19).   4:Calculate $\{l_{ai}^u,l_{bi}^u\}$ and $\{l_{aj}^v,l_{bj}^v\}$ from $H_{init}$ by Equation (16) and Equation (17).   5:Detecting salient line segments in the non-overlapping region to obtain $\left\{l_{ak}^s\right\}$.   6:Merge collinear local line segments into global lines $\left\{l_{ag}^s\right\}$.   7:Uniformly sample $\left\{l_{ai}^u\right\}$,$\left\{l_{ai}^v\right\}$,$\left\{l_{ak}^s\right\}$,$\left\{l_{ag}^s\right\}$.   8:Solve $\widehat{V}$ via minimizing the total energy term in Equation (26).   9:Warp $I_a$ via the bilinear interpolation on the basis of $\widehat{V}$.   10:Stitching the warped $I_a$ with $I_b$ via linear blending or seam-cutting blending.