2007-IJCV-Automatic Panoramic Image Stitching using Invariant Features资料.pdfVIP

Automatic Panoramic Image Stitching using Invariant Features Matthew Brown and David G. Lowe {mbrown|lowe}@cs.ubc.ca Department of Computer Science, University of British Columbia, Vancouver, Canada. Abstract can provide very accurate registration, but they require a close initialisation. Feature based registration does not re- This paper concerns the problem of fully automated quire initialisation, but traditional feature matching meth- panoramic image stitching. Though the 1D problem (single ods (e.g., correlation of image patches around Harris cor- axis of rotation) is well studied, 2D or multi-row stitching is ners [Har92, ST94]) lack the invariance properties needed more difficult. Previous approaches have used human input to enable reliable matching of arbitrary panoramic image or restrictions on the image sequence in order to establish sequences. matching images. In this work, we formulate stitching as a In this paper we describe an invariant feature based ap- multi-image matching problem, and use invariant local fea- proach to fully automatic panoramic image stitching. This tures to find matches between all of the images. Because of has several advantages over previous approaches. Firstly, this our method is insensitive to the ordering, orientation, our use of invariant features enables reliable matching of scale and illumination of the input images. It is also insen- panoramic image sequences despite rotation, zoom and illu- sitive to noise images that are