Login | Register

Deep Shape Representations for 3D Object Recognition

Title:

Deep Shape Representations for 3D Object Recognition

Ghodrati Asbfroushani, Hamed (2017) Deep Shape Representations for 3D Object Recognition. PhD thesis, Concordia University.

[thumbnail of Ghodrati_PhD_S2018.pdf]
Preview
Text (application/pdf)
Ghodrati_PhD_S2018.pdf - Accepted Version
Available under License Spectrum Terms of Access.
25MB

Abstract

Deep learning is a rapidly growing discipline that models high-level features in data as multilayered
neural networks. The recent trend toward deep neural networks has been driven, in large part, by
a combination of affordable computing hardware, open source software, and the availability of
pre-trained networks on large-scale datasets.
In this thesis, we propose deep learning approaches to 3D shape recognition using a multilevel
feature learning paradigm. We start by comprehensively reviewing recent shape descriptors,
including hand-crafted descriptors that are mostly developed in the spectral geometry setting and
also the ones obtained via learning-based methods. Then, we introduce novel multi-level feature
learning approaches using spectral graph wavelets, bag-of-features and deep learning. Low-level
features are first extracted from a 3D shape using spectral graph wavelets. Mid-level features are
then generated via the bag-of-features model by employing locality-constrained linear coding as a
feature coding method, in conjunction with the biharmonic distance and intrinsic spatial pyramid
matching in a bid to effectively measure the spatial relationship between each pair of the bag-offeature
descriptors.
For the task of 3D shape retrieval, high-level shape features are learned via a deep auto-encoder
on mid-level features. Then, we compare the deep learned descriptor of a query shape to the
descriptors of all shapes in the dataset using a dissimilarity measure for 3D shape retrieval. For the
task of 3D shape classification, mid-level features are represented as 2D images in order to be fed
into a pre-trained convolutional neural network to learn high-level features from the penultimate
fully-connected layer of the network. Finally, a multiclass support vector machine classifier is
trained on these deep learned descriptors, and the classification accuracy is subsequently computed.
The proposed 3D shape retrieval and classification approaches are evaluated on three standard 3D
shape benchmarks through extensive experiments, and the results show compelling superiority of
our approaches over state-of-the-art methods.

Divisions:Concordia University > Gina Cody School of Engineering and Computer Science > Concordia Institute for Information Systems Engineering
Item Type:Thesis (PhD)
Authors:Ghodrati Asbfroushani, Hamed
Institution:Concordia University
Degree Name:Ph. D.
Program:Information and Systems Engineering
Date:23 October 2017
Thesis Supervisor(s):Ben Hamza, Abdessamad
ID Code:983148
Deposited By: HAMED GHODRATI ASBFROUSHAN
Deposited On:05 Jun 2018 15:18
Last Modified:05 Jun 2018 15:18
All items in Spectrum are protected by copyright, with all rights reserved. The use of items is governed by Spectrum's terms of access.

Repository Staff Only: item control page

Downloads per month over past year

Research related to the current document (at the CORE website)
- Research related to the current document (at the CORE website)
Back to top Back to top