ArtWork recognition in 360-degree image using 32-hedron based rectilinear projection and scale invariant feature transform
Xun Jin, Jongweon Kim · 2016
Recently, with the increased generalization of 360-degree multimedia, the copyrighted art works are photographed without permissions. To prevent the infringements, we propose a method of art work recognition in 360-degree image using a 32-hedron based rectilinear projection and the well-known scale invariant feature transform (SIFT). The method transforms an equirectangular projected 360-degree image to a 3-dimensional sphere and generates a 32-hedron composed of 12 pentagons and 20 hexagons to wrap the sphere. It finds the nearest points on the sphere from vertices of the 32-hedron. The distances between these points are used to define the height and width of the rectilinear projection range. It also finds the nearest points on the sphere from the centers of 32 polygons. The locations of these points are used to define the viewpoint directions of rectilinear projection. With those parameters, the equirectangular projected 360-degree image is transformed to several rectilinear projected images. The recognition is implemented by matching the SIFT features extracted from the transformed images to those of original art work. The experimental results show some comparisons of matching results between before and after applying the proposed method of 32-hedron based rectilinear projection. With the proposed method, the features are well matched. Otherwise, the features can't be matched because of the seriously distorted equirectangular projection.