Quantitative Assessment Method of Image Stitching Performance Based on Estimation of Planar Parallax

While parallax-tolerant image stitching is a relatively mature field, the performances of image stitching methods have been assessed subjectively and qualitatively. These methods primarily provide the stitched image itself to demonstrate the performance, rather than quantitative data. Although several objective assessment methods have been proposed for quantifying the quality of stitched images, only the stitched output images have been analyzed, without considering the parallax level in each input image. We propose a method for quantifying the parallax level of the input images and clustering them accordingly. This facilitates a quantitative assessment of the various stitching methods for each parallax level. The parallax levels of the images are grouped based on the magnitude and variation in the planar parallax, as estimated with the proposed metric using matching errors and patch similarity. The existing image stitching methods are compared experimentally in terms of the residual misalignment errors, based on 73 pairs of different levels of parallax images originally classified in this study. Among the existing methods, the elastic local alignment method exhibits the least error. The shape-preserving half-projective method produces a larger misalignment error, but creates a natural panorama with less geometric distortion. We introduce a quantitative assessment method for considering the parallax of input images in image stitching methods. It can aid in specifying their performances, and in finding an appropriate method depending on the parallax level of the input images.

Automated summary

While parallax-tolerant image stitching is a mature field, the assessment of the performance of image stitching methods has been primarily subjective. This study proposes a method to quantify the parallax level of input images and cluster them accordingly for a quantitative assessment of stitching methods. Experimental comparisons show variations in residual misalignment errors among existing methods, with the elastic local alignment method exhibiting the least error and the shape-preserving half-projective method producing a larger misalignment error but less geometric distortion in the panorama.