Wide baseline image mosaicing by integrating MSER and Hessian-Affine
Yufang Ning, Ren Chen, Pengfei Xu · 2011
In this paper we propose a novel approach for wide-baseline image mosaicing which integrates MSER and Hessian-Affine detectors. MSER and Hessian-Affine are both robust detectors for wide-baseline stereo matching and they can be integrated owing to their availability in the structured scenes and the rich-textured scenes separately. However, the output shape of them is different, so they cannot be directly integrated. We use an affine covariant construction method to unify their output shape. At the same time, we introduce a standard elliptic equation to unify the ellipse parameters. The axial length and rotation matrix of ellipse with scale are calculated in accordance to the eigenvalue and eigenvector of image feature regions. Then MSER and Hessian-Affine regions are constructed as standard elliptical regions, and described as a unified parameter form. Our method provides more plentiful and robust features so that wide-baseline images can be stitched well. We design an experiment to compare the proposed method with the method based on SIFT. By testing 30 various image pairs, our experiment indicates that the proposed method is effective and available for the wide baseline images mosaicing, especially in the structured scenes with rich texture.