PhD Seminar: Video Quality Assessment: Exploring the Impact of Frame Rate

Wednesday, November 7, 2018 10:30 am - 10:30 am EST (GMT -05:00)

Candidate: Rasoul Mohammadi Nasiri

Title: Video Quality Assessment: Exploring the Impact of Frame Rate

Date: November 7, 2018

Time: 10:30 AM

Place: E5 4047

Supervisor(s): Wang, Zhou

Abstract:

Technology advancements in the past decades have led to an immense increase in data traffic over various networks. Videos constitute a major percentage of this traffic and their share is projected to increase at an accelerating speed in the coming years. Service providers aim to deliver videos that have high quality while at the same time keeping the data rate as low as possible. Effective and efficient objective Video Quality Assessment~(VQA) algorithms are essential in order to provide real time estimate of video quality so that the best compromise between data rate and quality can be achieved. Data rate of video transmission can be altered by controlling different parameters of the video, among which frame rate is one of the most important parameters. A good VQA algorithm should take into account the impact of altering different video parameters on video quality. So far, only limited works have been done to study the impact of frame rate variations on video quality.

The purpose of this work is to investigate the impact of varying frame rate on the quality of videos and to develop novel VQA models that integrate frame rate variations into the task of quality assessment. In order to achieve this goal, we first construct two new video databases that contain videos of diverse content, and spatial and temporal resolutions. We carry out subjective studies on these databases to obtain human opinions on video quality. The subjective study allow us to evaluate the performance of well known objective VQA algorithms on cross-frame rate videos. The results reveal that there is considerable disparity between the subjective scores and the predictions from state-of-the-art objective models that do not take frame rate into consideration. By making use of the subjective data collected from our experiment, we develop data-driven models for VQA using support vector regression (SVR) and feedforward artificial neural network (ANN) methods. We show that these simple regression models are effective at improving the quality prediction performance of existing VQA models that are ignorant of frame rate variation.

We also explore statistical models for video quality analysis. In particular, we conduct cross-frame local phase statistical analysis, which provides new insights on video motion smoothness, as an important factor that affects video quality across different frame rates. Our evaluations of proposed motion smoothness metric based on subjective-rated databases show that this novel measure provides a new means to capture the impact of frame rate on video quality, and demonstrates strong promise at improving the performance of objective video quality assessment models.