Content-based filtering for fast 3D reconstruction from unstructured web-based image data

Abstract

The huge amount of visual collections provides a unique opportunity for cultural heritage e-documentation and 3D reconstruction. The main difficulty, however, is its unstructured nature. In this paper a new content-based image filtering is proposed to discard image outliers that either confuse or significantly delay the 3D reconstruction process. The presented approach exploits a densebased unsupervised paradigm applied on multi-dimensional manifolds where images are represented as image points. The multidimensional scaling algorithm is adopted to relate the space of the image distances with the space of Gram matrices to compute the image coordinates. Evaluation on a dataset of about 31,000 cultural heritage images being retrieved from internet collections with many outliers indicate the robustness and cost effectiveness of the proposed method towards an affordable 3D reconstruction.