Biometric Identification with Limited Data Set

Sometimes, in real life, there is only limited information which can be obtained. This situation can be a challenge for the recognition system. The most suitable method to produce best performance needs to be developed. In this paper, small database with limited training data, 50 total images that were obtained from 25 people, with only 2 images for every respondent were tested in the recognition system. The SIFT algorithm was utilized and also compared with other methods such as Haar wavelet transform, principal component analysis and hierarchical Gaussian scale-space. The best recognition precision was produced by SIFT algorithm which was 38%, while other methods only gave out 30-32 % precision of recognition.