Local Directional Pattern (LDP) algorithm is a local feature extraction method which is robust to random noise. It has been widely used in face recognition. As LDP algorithm encodes the negative directions convolved by the Kirsch masks, the value of center gradation can’t fully reflect the information in the field of the local features. Thus face recognition performance will be influenced. A new algorithm named CLDP based on LDP algorithm is proposed. The CLDP algorithm encodes the positive directions which are convolved by the Kirsch masks. These directions can fully reflect the information in the field of the local features. And thus face recognition accuracy can be enhanced. To evaluate the performance of the proposed CLDP algorithm, some experiments are conducted in YALE, ORL and JAFFE face databases. Experimental results show that our CLDP algorithm has better recognition performance compared with LDP, LDN and ELDP algorithm.