On the relevance of image acquisition resolution for hand geometry identification based on MLP

Abstract

The effect of changing the image resolution over a biometric system based on hand geometry is analyzed in this paper. Image resolution is progressively diminished from an initial 120dpi resolution up to 24dpi. The robustness of the examined system is analyzed with 2 databases and two identifiers. The first database acquires the images of the hand underneath whereas the second database acquires the images over the hand. The first classifier identifies with a multiclass support vector machine whereas the second classifier identifies with a neural network with error correction output codes. The four experiments show that an image resolution of 72dpi offers a good trade-off between performance and image resolution for the 15 geometric features used.