Focusing on outcomes (a)–(c) (noted above), the objective for
this study was to quantify differences in image correspondence
between high quality exemplars and low quality known-matches
(KM) both pre- and post-filtering. In order to evaluate the
relatedness of two prints in a pairwise comparison, a traditional
similarity metric was implemented, capable of describing the
correspondence in minutiae identity and spatial configuration
between a questioned and exemplar print. To automate this
process, a two-step procedure was required that first registered the
fingerprints under comparison, and then evaluated the correspondences
assuming a simple point-pattern matching problem
[8–10]. Of course, the prealignment step was vital since knownmatch
images typically possess inherent variability. Among other
contributing factors, this mismatch between replicate images of a
print is a product of linear registration issues (translation, rotation
and uniform changes in scale), variations in image overlap,
variations in image quality/noise, and further complicated by
non-linear variables that are a function of uneven deposition
pressure and skin deformations [8]. These alignment problems are
so likely to exist that sophisticated solutions are required when
attempting to determine the appropriate registration transformation
that best describes known-match replicate images of the same
print. To mitigate these factors, several approaches to registration
have been proposed in the literature [11–16]. The process typically
begins with extraction of relevant print descriptors. These
descriptors are often detectable minutiae, but non-minutia
features such as ridge orientation fields or pores can also be
utilized [17]. Regardless of the features extracted for alignment
purposes, registration models must include, at minimum,translation and rotation [17]. However, more advanced treatments
of the registration problem attempt to mitigate the aforementioned
elastic deformations in an effort to increase the computed
similarity between known-match comparisons [10,15,16,18].