Adaptive image stitching algorithm based on optimized seam line

Image stitching has a wide range of applications in computer vision and graphics. It refers to the process of combining two or more images with spatially overlapping regions into a wide-angle panoramic image that contains complete information. In the process of stitching overlapping areas, ghosting or noticeable breaks often occur if there are moving objects or foreground elements. To achieve better quality in conventional panoramic images, we propose an image stitching method based on an optimized optimal seam line. The optimization of the seam line is achieved by selecting the path with the lowest energy, minimizing visual artifacts. This method uses a dynamic programming approach to select the optimal seam line by minimizing an energy function based on color, gradient, and the hue-saturation-intensity (HSI) color space. During path tracing, we expand the traditional path selection to cover additional columns, comparing the energy values of offset columns to ensure that a locally optimal path is selected. The reference image and target image are then stitched into a panorama according to the selected optimal seam line and the appropriate overlapping region, using a weighted fusion algorithm. Experimental results show that our method achieves superior stitching quality.