Abstract

This paper presents work in progress to extend the two-dimensional (2D) Scale Invariant Feature Transform (SIFT) to a 2.5 dimensional (2.5D) domain. Robust feature descriptors are extracted from range images of human faces and the form of these descriptors are analogous to the structure of Lowe’s 2D SIFT, in which the descriptors comprise a local distribution function of the image gradient orientations at each sampling keypoint location over a local support region. We adapt this concept into 2.5D domain by taking the relative frequencies of the surface types, derived using the [-1,1] bounded shape index, and their orientations into consideration, formulating a feature descriptor by applying a local support region over each keypoint landmark location. Nine sub-regions with the spatial support at one standard deviation are placed within the measurement sampling aperture in order to minimise the sampling effects caused by spatial aliasing. As a result, this feature descriptor is robust to viewpoint rotational changes. 1

Citations