A new optimal seam selection method for airborne image stitching

Image stitching techniques are widely used to integrate multi-view or sequential images into one image. But visible seams, blurring, or ghosting effects may occur in the overlapping regions for real image stitching. One approach to solve this problem is to find some kinds of seams in regions where the images agree, and then a weighted or blended integrating can be carried out around the pixels on the seams, which makes the transitions from one image to the other not visible. The traditional dynamic programming based method is proved to be effective to find such optimal seam but has an obvious drawback that its searching direction is restricted, so that it is not suitable to find an optimal seam connecting two specified points which is often required for sequential image stitching. Facing the above problem, a new optimal seam definition based on minimum cost is given, and then an optimal seam selection method based on enhanced dynamic programming is proposed in this paper. The analysis and visual results of our experiments using some real airborne sequential images show that, the approach proposed in this paper can handle significant structure misalignment and eliminates structure seam for the global seamless image stitching.