User Satisfaction-Driven Bandwidth Allocation for Image Transmission in a Crowded Environment

TL;DR

This paper introduces a user satisfaction-driven bandwidth allocation scheme for transmitting high-quality images in crowded environments, focusing on salient objects to optimize perceived quality and transmission delay.

Contribution

It proposes a novel, data-driven user satisfiability metric based on image saliency and machine learning, enhancing bandwidth allocation for better user experience.

Findings

The proposed metric correlates well with user satisfaction based on survey data.

Bandwidth allocation optimized by the new metric improves perceived image quality.

The approach outperforms fixed-function methods in crowded network scenarios.

Abstract

A major portion of postings on social networking sites constitute high quality digital images and videos. These images and videos require a fairly large amount of bandwidth during transmission. Accordingly, high quality image and video postings become a challenge for the network service provider, especially in a crowded environment where bandwidth is in high demand. In this paper we present a user satisfaction driven bandwidth allocation scheme for image transmission in such environments. In an image, there are always objects that stand out more than others. The reason behind some set of objects being more important in a scene is based on a number of visual, as well as, cognitive factors. Being motivated by the fact that user satisfaction is more dependent on the quality of these salient objects in an image than non-salient ones, we propose a quantifiable metric for measuring user-satisfiability (based on image quality and delay of transmission). The bandwidth allocation technique proposed thereafter, ensures that this user-satisfiability is maximized. Unlike the existing approaches that utilize some fixed set of non-linear functions for framing the user-satisfiability index, our metric is modelled over customer survey data, where the unknown parameters are trained with machine learning methods.