Fast object recognition using local scale-invariant features

Methods for generating local scale invariant features are introduced. These scale invariant features are used in object recognition. In order to improve the efficiency of object recognition, the feature points are detected and matched from large scale to small scale using hybrid multi-scale representation, which has integrated the advantages of pyramids and scale-space representation. This repeat process doesn't terminate until the recognition has completed. Experiment results show that this method has reduced the time of recognition effectively.