A novel approach to multi-resolution technique for fast pattern recognition

Multi-resolution is a commonly used acceleration algorithm for pattern search. However, there still exist two unsolved questions: "how many levels of image pyramid should be built" and "how to propagate the badly degraded truth patterns to the next pyramid level". We address these two questions through a novel distinction analysis and propagation strategy based on multimodality analysis. Experimental results show that this method has good practicability and advantage.