Index trees for accelerating deformable template matching

摘要

An improved method for deformable shape-based object detection and segmentation is described. A pre-computed index tree is used to improve the speed of deformable template fitting. Simple shape features are sued as keys in a pre- generated index tree of model instances. A coarse to fine indexing scheme is used at different levels of the tree to further improve speed. The index tree approach is demonstrated as an improvement to a previously-reported template-based region segmentation system. Experimental results show that when index trees are used, the speedup is significant while accuracy of shape-based indexing is maintained.