Find a SECReT supervisor
Information for overseas students
View SECReT animation
Download SECReT brochure

Use of a mirror-symmetry prior in small vehicle imaging

18 January 2011

Theo Mourouzis

Symmetry is a fundamental concept which can be used as a complexity-reducing tool that we find in many natural and man-made objects. The three types of symmetry; mirror, rotational and translational are directly related to shape perception. Formally, each type is defined as an invariant under some Euclidean transformation, such as reflection, rotation and translation, respectively.
The main aim of this project is to examine if the mirror-symmetry prior can be used successfully in small vehicle imaging for the detection of any suspicious items that are presented. As the initial problem involves a lot of complexities, we employ a simplified version of the problem, where the imaged subject  (P) is decomposed to (S)+(N), subject to different constraints, where (N) is a mirror-symmetric part and (S) is the part arising from the trivial subtraction (P)-(S) . The decomposition is done using the Simulated Annealing Algorithm (SAA) a heuristic probabilistic algorithm. Initially this idea was employed for the 3D reconstruction of a given structure from some 2D projection images, but then due to some computational difficulties we decided to limit our study to these specific 2D projection images.

Finally this idea is applied to four different images; one where a natural amount of mirror symmetry and a simple non-symmetric part are presented, one where the non-symmetric part is more complex and another two images where more complex mirror-symmetry is presented. From our results we conclude that the performance of the algorithm is higher if the images used have a large amount of mirror-symmetry, where the performance is compared relative to the Hamming distance. This is indeed a very promising result for the case of small vehicle inspection, as we expect cars to be approximately mirror-symmetric.