Practical Parameers Design for Gabor Filers with Application to Face Recognition

Feature extraction works using Gabor filters have shown that recognition performance is heavily affected by the filters parameters design. This paper investigates methods in this domain and proposes a practical one. The main contribution of this paper includes: (1) On the basis of the obvious directional characteristics in Gabor features, the filter-bank parameters design is transformed to a single filter optimization. The algorithm is simple and tractable in computing. (2) A model is presented to evaluate the Gabor features' discrimination based on the Fisher criterion. These methods are evaluated and compared on the NUST603 face database. Experimental results illustrate the availability and efficiency.