提出在Gabor 濾波理論的基礎上，結合Fisher 線性判別方法，對手寫數字圖像的所
有特征點尋找局部最優濾波頻率和濾波方向，從而提取最優Gabor 特征的方法。對MNIST 手寫體數據庫的識別實驗表明，該方法在小樣本情況下明顯優于直接Gabor 特征提取。
關鍵詞：Gabor 濾波；Fisher 線性判別；手寫數字識別；特征提取；LIBSVM

Extract Gabor Features Based on Fisher’s Discriminant ZHOU hui-can1, LIU qiong1，WANG yao-nan2 (1.Hunan University of Art and Science, Department of Computer Science, Changde Hunan 415000，China； 2.Hunan University, College of Electrical and Information Engineering, Changsha Hunan 410082，China) Abstract：A method for extracting optimal Gabor features is presented in this paper, witch bases on the theory of Gabor filters and combines the Fisher’s linear discriminant. In an image of handwritten numeral, the above-mentioned method tries to find the local optimal frequencies and orientations for filtering at all of the feature points. An experiment of handwritten numeral recognition to MNIST database indicates that the method has an advantage of directly extracting Gabor features conditions of small sample obviously.
Key words：Gabor filters；Fisher’s linear discriminant；Handwritten numeral recognition；Features extracting；LIBSVM