The invention provides a character recognition device and a method. The character recognition device comprises a feature vector extraction unit which extracts feature vectors of a character image from a training character image or a character image to be recognized; a training unit for carrying out study according to a training sample set which is composed of the feature vectors of the training character image and output by the feature vector extraction unit, thereby obtaining a plurality of sparse support vector machine classifiers; a storage unit for storing a training result of the training unit; and a recognition unit for calculating outputs of various sparse support vector machine classifiers to the feature vectors of the character image which is ready to be recognized and output by the feature vector extraction unit, thereby determining a character which is corresponding to the character image to be recognized; wherein, the training unit reduces the number of support vectors in decision functions of the sparse support vector machine classifiers by introducing 0-norm regularization items into target functions of the sparse support vector machine classifiers.