Transformation-variants estimation using similarity relative histogram grouping model

This paper proposes a novel method that estimates parameters of a geometric transformation by means of grouping and statistical inference in relative histogram grouping models borrowed from the law of large number. Since an image is divided into many local partial objects, each of which is represented by groups of transformation-invariant and transformation-variant (e.g., translation, rotation, and shearing) and whose deformations are considered to be linear, we construct minutiae-simplexes to represent partial objects in a fingerprint. A relative histogram grouping model describes the relationship between transformation-invariant and variant. Even if the image is transformed at random, partial objects still maintain their linear representation, from which the relative histogram techniques extract grouping centers that account for transformation-variations in partial objects for geometric alignment. Our promising experimental results show that our technique is efficient in alignment and match.