Abstract: Most of existing writer identification methods usually require the samples of strict format or many handwriting characters (average more than 150 characters). However, these conditions cannot be always attained in practical applications. When the samples have fewer characters and looser handwriting pattern conditions, these existing methods have lower identification performances. Aiming at the above shortcoming, this paper proposes an adjacent ring structure (ARS) feature algorithm. The reason is introduced for utilizing the principal component analysis (PCA) and linear discriminant analysis (LDA) method and the working procedure of deep belief network (DBN) is stated. The performance comparisons from different aspects are made. In the proposed identification method, the first step is to preprocess the handwriting Chinese character images by taking a random sample of handwriting contour images to get patches of the same size. And then, the proposed ARS algorithm is used on the patches for extracting features whose multiple patch features represent the stylistic information of one writer. Finally, both PCA and LDA are utilized to reduce the feature dimensions so that the dimensionality curse can be avoided. Besides, DBN is used to train the identification models of different writers and count the correct identification rate. This proposed method is text-independent, simple, and easily realized. On the small samples with average 45 Chinese characters per sample, the proposed method can still effectively represent the stylistic information of different writers. It is shown from the experiments on HIT-MW handwriting identification database that the proposed method can achieve similar performance to other identification methods using large amount of characters.