Abstract: A new subspace algorithm(LSM-ALSM) is discussed in this paper. By rotating the subspaces of ALSM, LSM changes their independent relations and improves the ALSM classifier's recognition capacity in alike characters. To realize the best balance between recognized rate, rejected rate and error rate, we presents performance function instead of experience function to determine the refuse rules. The experiment with limited character set shows that LSM-ALSM method improves the recognized rate and credibility effectively.