Remote Video Quality Assessment Using Machine Learning

Video is a crucial component of communication in the modern era with applications in every field. The use of mobile technology has enabled anyone to create videos, however, this has resulted in a range of video quality scenarios. As a result of this, assessing the quality of these videos has become an important area of research. Over time various metrics were defined and used to measure the quality including classical metrics such as SSIM and PSNR. These studies have become more advanced in recent years, with the emergence of machine learning-based approaches. Machine learning-based approaches have been implemented as a way to measure VQA (Video Quality Assessment). Such machine learning-based techniques are proving to be effective in accurately measuring video quality. Thus, helping ensure that effective improvement feedback is provided in case of online video communication in real-time. This research study outlines an effective approach using two machine learning-based stages to derive the VQA metric. The research study employs the feature extraction process in the first stage using the pre-trained ResNet50 and ResNet152 models to extract the features. The second stage makes use of the MOS—Mean Opinion Score and combines it with the temporal pooling methodology to arrive at the final VQA metric. The results section provides a comprehensive comparison of machine learning parameters such as KROCC, and SROCC comparing the feature extraction pre-trained models used in the study. The validation outcomes of KROCC and SROCC demonstrated that the ResNet152 model performed marginally better than the ResNet50 model, as demonstrated by the experiment results. This indicates the efficiency of the proposed machine learning model in accurately assessing the quality of movies and highlights the significance of selecting models for feature extraction in machine learning applications with care.