特征提取是手寫體數字識別研究中的重要問題，有效、穩定的特征是提高識別率和識別精度的關鍵。該文提出了一種基于分數本征特征和核非線性分類器的手寫數字識別方法，首先找到時頻平面的一個軸進行分數傅里葉變換，使不同類別樣本在這個軸上最大限度地分開，然后用主元分析進行降維，得到比較穩健的低維特征，再將常用分類器用于特征分類，實現對手寫數字的識別。對實際數據進行實驗，結果表明上述本征特征與核非線性分類器相結合有較高的識別率和訓練、分類效率。Feature extraction is an important part in handwritten numeral recognition. Efficient and robust feature is a key to improving recognition rate and efficiency. This paper adopts fractional Fourier transform and principal component analysis to extract robust and compact features called fractional eigenafeatures. In classification, five kernel-based nonlinear classifiers, Parzen and robust Parzen classifiers, radial basis function classifier, support vector classifier, and kernel-based nonlinear representor are applied and compared. Experimental results show the effects and efficiency of the proposed algorithm.