Matching degraded partial palmprint images against full palmprints is a challenging problem, since these images may be arbitrarily rotated, incomplete and often noisy. Such partial palmprints can be recovered from the palm impressions left on some surface (called latent partial palmprints) or can be generated, for testing purposes, by cropping full palmprints into different regions/segments (called synthetic/ pseudo latent partial palmprints).
This paper proposes a new technique, PP-RIDER – Partial Palmprint Rotation-Invariant and DEgraded Recognition, for recognizing degraded partial palmprints, which combines the Fourier-Mellin Transform (FMT) with the Modified Phase-Only Correlation (MPOC) technique. FMT is used to correct the arbitrary rotation of partial palmprints. Then, the concept behind MPOC is used for matching the degraded, but aligned, partial palmprint to a full palmprint registered in a database. Experimental results, using the THUMPALMLAB high resolution palmprint database, from which partial palmprints were cropped, randomly rotated and further degraded by adding white additive Gaussian noise and motion blur, show an improvement in comparison to the original MPOC technique.