Abstract:
To present a fast biometric personal identification method. Instead of matching an input image feature with all templates in the database as in general identification systems, this research proposes to match it only with those that have similar palmprints. Palmprints are classified based on three principle lines which are a life line, a head line and a heart line because they are obvious and unique. The feature used for identification is extracted from a palmprint with Log-Gabor filter. With the proposed palmprint identification method, a palmprint image is taken from an unfixed position of a hand. After finding the palm area or a region of interest, three principle lines are extracted. Plamprints in this research are classified into several groups based on their characteristics. The palmprints Log-Gabor filtered feature is matched with those of the same group. Hamming distance is used for evaluating the feature similarity. If the matching result is not found within the group, it continues to the next closest group(s). Experiments are done with three palmprint databases. They are Visgraph [8] database (the palmprint images of Hong Kong peoples), CU-CGCI1-Hand database (those of Chulalongkorn University students) and CU-CGCI2-Hand database (those of Thai people with different occupations). It is found that the distribution of the six groups from palmprint classification conform in all databases. The distribution from the largest group to the smallest group is about 34%, 3.4%. For personal identification tested with all three databases, the Equal Error Rates are around 3.22%, which is comparable with the traditional identification process of 3.1%, but the number of feature matching is greatly reduced to approximately 40%.