Seamless Image Stitching Using Optimized Boundary Matching for Gradient and Curvature

We propose an image stitching method which can compose the images with similar overlapped regions seamlessly. Traditional minimal error boundary (MEB) based method can´t produce correct stitching result when the input images exist photometric inconsistency. To cope with the problem, we adopt the weighted sum of gradient and curvature as the robust feature, which represent the primary content of the images and are insensitive to photometric change and exposure difference, to match the overlapped regions. The optimal boundary can be found and the pixels of stitched image are taken from the input images in corresponding side of the boundary to form satisfactory result. Experiments are carried out to show effectiveness of the method.