Content dependent spatial resolution selection for MPEG DASH segmentation

Abstract

Abstract The need for standardization emerged with the fast development of video streaming technologies. MPEG Dynamic Adaptive Streaming over Hypertext Transfer Protocol (MPEG DASH) is a standard for adaptive video streaming. Applications developed in compliance with MPEG DASH ensure smooth playback to end-users while being hardware agnostic, cost-efficient, and easy to maintain. The client side of the MPEG DASH system houses the adaptation logic which has been investigated intensively over the last several years. The server side stores representation sets of video sequences as well as accompanying Media Presentation Description and initialization file. Compared to the client side, the server side has not been yet sufficiently explored. Although there are some papers investigating the parameter selection for representation sets, most solutions are proprietary, do not take into account the content, and do not give the comprehensive methodology and its notation that can be applied to other codecs and encoder parameters. Also, most of the currently available solutions are costly and require large computational power. This article provides a comprehensive analysis and methodology for selecting the most appropriate spatial resolution switching points, which define the bitrates where switching to the higher spatial resolution is optimal. Furthermore, it presents a spatial resolution selection model for the estimation of switching points, which considers the spatial and temporal activity of video sequences. With the developed model, the optimal representation selection process is simplified and the segmentation process in MPEG DASH is improved. The methodology proposed for the development of this content-dependent spatial resolution selection model can be used for encoders and coding parameters not covered by the presented research. Quality of the video sequences segmentation using the proposed switching points were tested using two different test cases for network conditions and two different adaptation algorithms: Basic Adaptation algorithm and Segment Aware Rate Adaptation algorithm. The streamed videos that used the proposed segmentation have achieved better SSIM (Structural Similarity Index) results in 83.33% of cases in comparison to the videos that used segmentation presented in the relevant literature.