OBJECTIVE METHODS FOR VIDEO QUALITY ASSESSMENT USING DEEP LEARNING

Abstract

The objective methods of video quality evaluation without references make possible the identification of degradations and noises contained in videos, which are produced by several external factors present in the context of different telecommunication processes: acquisition, compression, processing, or transmission. Identifying degradations in the video is essential to improve its quality. Other processes need to be carried out later, such as detection, recognition of objects, or simply reproducing videos with higher quality. The methods of video quality assessment have evolved from the stage in which it was performed only from the perceptual or subjective evaluation of the quality based on the experience of experts. At the current stage, objective methods predominate using artificial intelligence techniques that perform the analysis from machine learning and are designed using convolutional neural networks. These processes are bio-inspired since they emulate how the human brain identifies noises and distortions in video sequences. This paper summarizes the study of the state of scientific knowledge in the field of video quality assessment methods based on deep learning, which has been developed in the last five years by the international scientific community. We aim to expand the potential for executing the process when large volumes of audiovisual information need to be managed in real-time.