PROJECT TITLE :

Scale and Rotation Invariant Matching Using Linearly Augmented Trees

ABSTRACT:

We propose a completely unique linearly augmented tree technique for economical scale and rotation invariant object matching. The proposed method enforces pairwise matching consistency outlined on trees, and high-order constraints on all the sites of a template. The pairwise constraints admit arbitrary metrics while the high-order constraints use L1 norms and thus can be linearized. Such a linearly augmented tree formulation introduces hyperedges and loops into the fundamental tree structure. However, different from a general loopy graph, its special structure permits us to relax and decompose the optimization into a sequence of tree matching issues that are efficiently solvable by dynamic programming. The proposed technique also works on continuous scale and rotation parameters; we tend to can match with a scale up to any large price with the identical efficiency. Our experiments on ground truth knowledge and a selection of real pictures and videos show that the proposed technique is efficient, accurate and reliable.