PhD Oral Exam - Muhammad Reza Pourshahabi, Electrical and Computer Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Robotic-assisted minimally invasive surgery (MIS) offers numerous benefits including smaller incisions, faster recovery, enhanced precision, and remote operations. Image processing operations such as 3D visualization, augmented reality, and image registration, which are often feature-based, are used in MIS. Feature detection, extraction, and matching (FDEM) and feature matching refinement (FMR) constitute the cornerstone of these operations. MIS images are affected by deformation, occlusions, and specular reflection, which hinder the processes of FDEM and FMR, severely affecting the number of matched features.

FDEM is a process in which, given a pair of images, certain distinctive features are detected from the pair, then suitably represented as feature vectors, and finally, the corresponding feature vectors are compared and matched leading to a set of matched features known as a putative set for the pair. On the other hand, FMR is a process in which the falsely matched pairs of features are, as much as possible, removed from a putative set. The existing FDEM and FMR schemes are computationally expensive or lead to a set of matched features that are not well dispersed over the region of interest and suffer from having an insufficient number of true matches.

The overall objective of this thesis is to propose robust and fast schemes for generation of matched features in MIS images. In the first part of the thesis, a very fast and accurate FMR scheme is proposed. The main idea used in developing this scheme is in determining the size of local neighborhoods so that the smoothness of deformation field can be effectively applied to check the feature topology preservation between the corresponding regions of the pair of images to identify the true matches in the putative set of the pair. In the second part, a fast and accurate FDEM scheme that combines the good characteristics of three well-known FDEM schemes, SIFT, SURF and ORB, is proposed. The focus is on producing putative sets of matched features that have a good spatial quality in addition to a good matching quality. Extensive experiments are conducted to demonstrate the effectiveness of the proposed FMR and FDEM schemes.