Benchmark Study of Image Quality with Saliency-Guided Compression in Mobile Streaming Platforms

Abstract

The exponential proliferation of mobile streaming platforms has precipitated an unprecedented demand for efficient image and video compression techniques that can operate effectively under constrained bandwidth conditions while preserving the optimal human visual experience. Traditional compression algorithms often apply uniform degradation across the entire visual frame, fundamentally ignoring the physiological and psychological realities of human visual perception, wherein certain regions inherently attract disproportionate gaze fixation. Consequently, saliency guided compression has emerged as a promising paradigm, dynamically allocating bitrates to regions of high visual interest at the expense of peripheral background elements. However, the subsequent evaluation of visual fidelity in such heterogeneously compressed media presents a formidable challenge to conventional objective quality assessment metrics. This paper comprehensively investigates the prediction of image quality derived from saliency guided compression methodologies through a large scale benchmark study specifically tailored for mobile streaming environments. By constructing an extensive dataset encompassing a diverse array of source images subjected to spatially variant quantization, this research systematically evaluates the correlation between subjective human mean opinion scores and an array of sophisticated objective quality predictors. The primary objective is to elucidate the mechanisms by which saliency weighting influences perceived degradation and to establish an optimized algorithmic framework for predicting end user satisfaction. Extensive empirical analysis reveals that traditional full reference metrics consistently miscalculate the perceptual severity of peripheral artifacts, whereas newly proposed saliency integrated predictive models exhibit remarkably higher statistical correlation with human perception. Ultimately, this research provides vital theoretical foundations and practical guidelines for optimizing mobile media delivery architectures.