Abstract
This paper presents Slime, a novel non-deep image matching framework whichmodels the scene as rough local overlapping planes. This intermediaterepresentation sits in-between the local affine approximation of the keypointpatches and the global matching based on both spatial and similarityconstraints, providing a progressive pruning of the correspondences, as planesare easier to handle with respect to general scenes. Slime decomposes the images into overlapping regions at different scales andcomputes loose planar homographies. Planes are mutually extended by compatiblematches and the images are split into fixed tiles, with only the besthomographies retained for each pair of tiles. Stable matches are identifiedaccording to the consensus of the admissible stereo configurations provided bypairwise homographies. Within tiles, the rough planes are then merged accordingto their overlap in terms of matches and further consistent correspondences areextracted. The whole process only involves homography constraints. As a result, both thecoverage and the stability of correct matches over the scene are amplified,together with the ability to spot matches in challenging scenes, allowingtraditional hybrid matching pipelines to make up lost ground against recentend-to-end deep matching methods. In addition, the paper gives a thorough comparative analysis of recentstate-of-the-art in image matching represented by end-to-end deep networks andhybrid pipelines. The evaluation considers both planar and non-planar scenes,taking into account critical and challenging scenarios including abrupttemporal image changes and strong variations in relative image rotations.According to this analysis, although the impressive progress done in thisfield, there is still a wide room for improvements to be investigated in futureresearch.