Abstract: The long time span invariable feature(LTSIF)and short time span invariable feature(STSIF)exist in face features. A method was proposed to classify the two types of features， then the LTSIF was used to recognize the long time span face image. First， the bestdivide method was replaced by the bestcoverage method to normalize the features based on shuffled frog leaping algorithm(SFLA). Second， the normalized featureweight and basevalue was calculated by the time span and the featurevariability to classify the LTSIF and STSIF. In the recognition process， the featureweight and basevalue were refreshed with the recognized face image to recognize the long time span face images. The results indicated that over 82% face images can be recognized by the proposed method in years’ time span， which was better than other algorithms.

Key words: computer application, SFLA(shuffled frog leaping algorithm), feature normalization, face recognition, SIFT(scale invariant feature transform)