Abstract:

A new face recognition algorithm based on fusion local monogenic orientation,different phase scales information and the monogenic phase local difference binary pattern in the same scale is proposed. Firstly, we calculate the monogenic phase by the monogenic filters, and extract the local difference on monogenic phase, then obtain the absolute values from local difference and encode the results using binary patterns. Secondly, we encode the monogenic orientations and the same location points in every monogenic phase scale by using binary patterns, and then concatenate the monogenic orientation codes, different scales codes and the local difference codes of monogenic phase into a long binary string so as to form the monogenic phase local difference patterns.Finally,in order to remain the most information and reduce the feature dimensions, we form the mapping table with a statistical method, and map the original monogenic phase local difference patterns. Experimental results on ORL and CASPEAL databases show that,the proposed algorithm can achieve a better face  recognition effect ampared with stateofart methods. 

Key words: monogenic signal;phase;difference;binary code;statistical patterns