S. Liu (PRC)
Dynamic programming, shape matching, Multi-scale space
This paper presented an approach based on multiscale representation and Dynamic Programming for matching deformed and possibly occluded shapes, which is robust with respect to noise and invariant to scale, translation, orientation and starting point selection. The process of contour segmentation can adjust automatically while the amounts of noise and deformation change. And the correspondence of similar parts of shapes helps to analyze the object structure and can be used as prior knowledge to learn shape model. We have tested and evaluated our method on a database of 1100 images of marine animals with a vast variety of shapes with very good results.
Important Links:
Go Back
