[Defense] Image Quality for Object Detection in Compressed Videos
Friday, July 15, 2022
11:00 am - 12:00 pm
In
Partial
Fulfillment
of
the
Requirements
for
the
Degree
of
Doctor
of
Philosophy
Poonam
Beniwal
will
defend
her
dissertation
Image
Quality
for
Object
Detection
in
Compressed
Videos
Abstract
The amount of video data generated daily is enormous, making it nearly impossible for humans to understand the content of the data. The use of machine learning and deep learning approaches for automatic analysis has grown. It is crucial to determine the robustness and reliability of the automated analysis. The reliability of automated systems can be evaluated using a variety of factors. One such parameter is compression, which is an inherent part in video transmission and storage. We analyzed the impact of compression on three computer vision algorithms. The dataset used for analysis is collected from an IP-based surveillance camera and compressed using different bandwidths and quantization levels. We also find a correlation between image quality and the performance of algorithms. Existing image quality metrics cannot explain the drop in performance for object detection. The traditional image quality metrics define quality from a human perspective. We introduced full-reference and no-reference metrics to overcome the constraints of existing image quality metrics. We present the performance of the image quality metric on different aspects of object detection.
Friday,
July
15,
2022
11:00AM
-
12:00PM
CT
Online
via
Dr. Shishir Shah, dissertation advisor
Faculty, students and the general public are invited.
