Dissertation Defense - University of Houston
Skip to main content

Dissertation Defense

In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

Arko Barman

will defend his dissertation

Consensus-Based Decision-Making for Improving Person Re-Identification & Similar Systems


Person re-identification is a problem in video-based surveillance that deals with identifying the same person in multiple different camera views. The problem is particularly challenging since the images of persons obtained from a surveillance camera often suffer from a plethora of difficulties -- changes in illumination, viewpoint and pose, occlusions, and differences between indoor and outdoor settings. A number of algorithms have been proposed for solving person re-identification. The most popular approach towards solving the problem is to extract features from an image of a person (known as the ``probe image'') and to compute the ``distance'' of these features from the features extracted from a set of gallery images. These distances or similarity scores are then used for ranking the gallery images in terms of their similarity with the probe. Not only does the process of feature extraction vary widely, but the distance computation also utilizes a number of different distance metrics. These distance metrics may be pre-existing or trained using a set of annotated person images. Consequently, the distances or similarity scores which are used to rank the gallery images belong to different distributions and are of a widely varying range of values. In this dissertation, a number of algorithms for making consensus-based decisions using multiple person re-identification algorithms are proposed. The consensus-based decisions result in a significant improvement in performance over individual algorithms. The proposed algorithms are unsupervised and do not require any tuning depending on the datasets used or the individual algorithms used for computing similarity scores. Further, the proposed algorithms treat individual person re-identification techniques as ``black-box'' methods. In other words, no prior knowledge of how the individual algorithms work is assumed. This lends the property of generalization to the algorithms and allows them to be used not only for person re-identification but also for any retrieval or ranking problem in general. Finally, the fact that the proposed algorithms can be used in similar problems has been illustrated through its application to image search.

Date: Monday, April 9, 2018
Time: 9:00 AM
Place: PGH 501D
Advisor: Dr. Shishir Shah

Faculty, students, and the general public are invited.