Palm-vein recognition is among the more secure biometrics precisely because the pattern is under the skin. It cannot be lifted from a glass or captured from a photograph the way a fingerprint or a face can, and it needs a living hand present.

The difficulty is at the sensor. Near-infrared light scatters as it passes through tissue, and sensor limitations compound the problem, so the captured image tends to be low contrast: the pattern is there, faintly, mixed into surrounding noise. Every recognition method downstream inherits that weakness, however clever it is.

This work attacks the image rather than the classifier, proposing an adaptive contrast stretching method using bidirectional Gaussian-weighted overlapping tiles. Treating tiles with overlap and weighting matters because a naive per-tile enhancement produces visible seams, and a seam is precisely the kind of artefact a feature matcher will happily mistake for a vein.