Modifying and generalising the Radon transform for improved curve-sensitive feature extraction

[摘要] ENGLISH ABSTRACT : In this thesis a novel and generic feature extraction protocol that is based onthe well-known standard discrete Radon transform (SDRT) is presented. TheSDRT is traditionally associated with computerised tomography and involvesthe calculation of projection profiles of an image from a finite set of angles.Although the SDRT has already been successfully employed for the purposeof feature extraction, it is limited to the detection of straight lines.The proposed feature extraction protocol is based on modifications to theSDRT that facilitate the detection of not only straight lines, but also curvedlines (with various curvatures), as well as textural information. This is madepossible by first constructing a novel appropriately normalised multiresolutionpolar transform (MPT) of the image in question. The origin of said MPTmay be adjusted according to the type of features targeted. The SDRT, orthe novel modified discrete Radon transform (MDRT) conceptualised in thisthesis, is subsequently applied to the MPT.The extraction of textural information based on different textural periodicitiesis facilitated by considering different projection angles associated withthe MDRT, while the extraction of textural information based on differenttextural orientations is facilitated by specifying different origins for the MPT.The extraction of information pertaining to curved lines is made possible byspecifying origins for the MPT that are located at different distances from theedge of the image in question – the SDRT is subsequently applied to a givenMPT from a specific angle of 90 .An existing system that only employs SDRT-based features constitutes abenchmark. Two novel texture-based systems, that target different texturalperiodicities and orientations respectively, are developed. A novel system,that constitutes a generalisation of the SDRT-based benchmark, and is gearedtowards the detection of different curved lines, is also developed.The proficiency of the proposed systems is gauged by considering a dataset that contains authentic handwritten signature images and skilled forgeriesassociated with 51 writers. All of the proposed systems outperform the SDRTbasedbenchmark. The improvement in proficiency associated with each individualtexture-based system is statistically significant. The proficiency of theproposed systems also compares favourably with that of existing state-of-theartsystems within the context of offline signature verification.

[发布机构] Stellenbosch University