A dorsal hand vein pattern recognition algorithm

An algorithm for dorsal hand vein pattern recognition is presented in the paper. To improve the recognition ratio, the vein skeleton extracting with little distortion is very important. Firstly, the algorithm acquires a clean, one-pixel-width skeleton with little distortion after a series of processes: size and gray normalizing, Gaussian lowpass and median filtering, NiBlack local dynamic thresholding segmenting, area thresholding, morphological opening and closing, median filtering, conditional thinning, spurs pruning. Then, the seven corrected moment invariants of the vein skeleton are extracted as the feature vector. At last, the feature vector is input into SVM for training and recognition. Experiment shows the algorithm achieves a higher recognition ratio of 95.5%.