BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260722T154054EDT-3433WF8Z7f@132.216.98.100
DTSTAMP:20260722T194054Z
DESCRIPTION:Abstract\n\n \n\nTexture-flow are locally dense parallel patter
 ns common in natural images. The texture-flow structure of an image is the
  trace of intrinsic properties of the objects in a scene\, and understandi
 ng it is a crucial step in the human perception of an image. High-level pe
 rceptual tasks involving recognition\, detection\, classification and segm
 entation also rely upon the notion of texture-flow. In the past decades\, 
 researchers have proposed several computer vision methods for estimating t
 he texture-flow profile of an image. The current state-of-the-art techniqu
 es\, such as tensor-voting and relaxation-labeling come short of processin
 g sparse input images and cannot effectively manipulate scale. This thesis
  aims to provide a method for calculating the global texture-flow profile 
 of sparse images and deal with the scale variation of the edge patterns in
  a natural image. In this thesis\, firstly\, the new angular orientation p
 robability distribution function (AOPDF) is proposed for representing the 
 texture-flow profile of a natural image. At each location in the image\, A
 OPDF depicts the likelihood of the texture-flow and curvilinear angular-or
 ientations defined by a new two-parametric spatial angular orientation (AO
 ) function. Subsequently\, a new numerical method is proposed for estimati
 ng the AOPDF in digital images at discrete locations (pixels) and for a di
 screte set of angular-orientations. It is shown that AOPDF improves the re
 sults significantly when used for solving the challenging problem of singl
 e-image super-resolution. Furthermore\, the AOPDF is combined with an anti
 -aliasing filter and reformulated into a kernel form\, the so-called angul
 ar orientation of the edges (AngOri). The multi-scale AngOri kernels are i
 ntended to initialize the convolutional layers in deep neural networks (DN
 N)\, seamlessly. In a set of experiments\, the neural networks are trained
  for image segmentation and object detection tasks. In all cases\, the tra
 ined DNNs\, initialized with AngOri kernels\, achieve higher validation ac
 curacies\, especially when the training set is sparse. Finally\, a new ima
 ge reconstruction method is proposed for very sparse images\, where the te
 xture-flow could not be estimated from the image due to severe sparsity. T
 he missing parts of the image are constructed from a set of patterns chose
 n\, based on the similarity of local high-order statistics\, from the trai
 ning set. The image reconstruction results are significantly improved comp
 ared to the results obtained from existing methods. The new AOPDF\, AngOri
 \, and high-order stochastic methods introduced in this thesis are reliabl
 e alternatives for solving challenging computer vision tasks involving eit
 her sparse input images or training data.\n
DTSTART:20231024T140000Z
DTEND:20231024T160000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Amir Abbas Haji Abolhassani – A new probabilistic mo
 del for representation of the texture-flow in natural images
URL:https://www.mcgill.ca/ece/channels/event/phd-defence-amir-abbas-haji-ab
 olhassani-new-probabilistic-model-representation-texture-flow-natural-3521
 44
END:VEVENT
END:VCALENDAR