Hand Palm Recognition Using Wavelet Transform

Biometric identification is an emerging technology that can solve security problems in our networked society. Biometrics traits such as fingerprints, hand geometry, face and voice verification provide a reliable alternative for identity verification / identification and are gaining commercial and high user acceptability rate. In this research, a hand palm identification system has been introduced. The work flow of the introduced system consists of two stages. The first stage is called the preprocessing stage and the second stage is called matching stage. In the preprocessing stage a number of color images samples of the right hand for some persons have been captured. In both phases (preprocessing and matching)each color input image was converted to binary image by using thersholding method, then the boundary image extracted by using Laplace edge detector. In order to extract the hand palm area the sequence of the hand boundary point was collected by using chain coding algorithm. The last step in the preprocessing stage is the determination of the joint points which are required to separate the palm region from the overall hand area. After the joint points have been determined then Bresenham algorithm was implemented to connect the joint points by lines. Then seed filling algorithm are used to mark the palm pixels as white colored pixels, and finally the white color of the marked pixels was replaced with the original color of these pixels, the output of this stage was color palm region. Haar wavelet transform had been applied to determine the Haar wavelet coefficients. Then features of the palm area were extracted by using two types of discriminating features (i.e., complex moment and invariant moment which are stored as feature vectors in a database file. Also,the mean and standard deviation of the feature vectors had been determined to be used as template in the matching stage. In the matching stage the identification process was based on using Euclidean distance measure as a similarity criterion. The test results indicate that the identification rates when using complex moments features as discriminating features are much better than when using invariant moments features, and the best identification rates when using one pass Haar wavelet transform have reached 88% using complex moments features and 78% using invariant moments features, but when using two pass Haar wavelet transform the best identification rate have reached 96% when using complex moments features and 76% when using invariant moments features.