Feature extraction, as one of the two important components in handwritten digit recognition systems, is still a key research area. Principal Component Analysis (PCA) is an efficient linear feature extraction algorithm and is widely used in handwritten digit recognition system. However, it can hardly deal with the pattern with complex nonlinear variations, such as the writing interrupt, noise pollution and so on. This paper proposes an efficient handwritten digit recognition method based on distance Kernel PCA (KPCA). First, the initial input data is mapped into a higher-dimensional space with the distance kernel and describes the whole features as much as possible. Then, PCA method is used to extract the Principal Component from the kernel matrix. Last, SVM acts as the classifier to make decision. To test and evaluate the proposed method performance, a series of studies has been conducted on the MINST database. Compared with the other models, the approach proposed shows a better recognition rate and is more satisfying.
展开▼
