False mapped feature removal in spin images based 3D ear recognition

摘要

This paper proposes a methodology for 3D ear recognition using spin images and removing false mapped features using geometric surface properties of 3D shapes. Many Feature detection techniques works well with different 3D objects for recognition but not well with ear data of similar shapes. Since, ear data of the different subjects varies with unique features but fundamentally of similar shape. Detected features belongs to flat region of test ear may match with the features belongs to flat region of target data with different co-ordinate position, leads to cross feature matching. Using geometrical surface properties and extended neighbourhood of detected feature points, cross matched features has been removed, which increased the recognition rate. The experimental results carried out with UND Ear database gives a promising results.