CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: A point process model for pulses of the DPT from an image Authors:  Inger Fabris-Rotelli - University of Pretoria (South Africa) [presenting]
Alfred Stein - University of Twente (Netherlands)
Abstract: The Discrete Pulse Transform (DPT) is a multiscale decomposition of an image obtained using the LULU smoothers. The LULU smoothers are nonlinear and have many powerful properties as such edge and total variation preservation. They have been developed for multiple dimensions here we focus on images. A point pattern is extracted from an image using the DPT and modelled as a spatially clustered pattern with trend in both the rows and columns of the image. The clustered nature arises naturally from the images structures which are well extracted by the DPT, and is confirmed prior to modelling. The fitted models perform well and extracts the spatial intensity of the image. The DPT provides a representation of an image at all scales from detail up to background shading - and enables informed analysis of the content and features of an image. The aim is to expand the DPT image application into the spatial domain, a natural domain for image structures having obvious spatial dependence.