Adaptive Intra Period Size for Deep Learning-Based Screen Content Video Coding

Abstract

In recent years, there has been a significant upsurge in screen content video (SCV) data, which includes video content produced via the interactive interfaces of terminal devices and computer graphics. Alongside this growth, there has been a swift evolution in video coding algorithms powered by deep learning. However, traditional algorithms, mainly tailored for videos of natural scenes, often struggle to efficiently process the distinct sharp edges and high contrast features characteristic of SCV. In response to this challenge, we introduce an innovative deep learning-based strategy to enhance the efficiency of screen content video coding. Our approach in-volves dynamically adjusting the Intra-Period Size (IPS) between two intra-coding frames within the sequence, thereby optimizing the bitrate. Extensive experimental results have underscored the efficacy of our method, showcasing its potential to significantly improve the performance of screen content video coding. Related datasets can be found at https://openi.pcl.ac.cn/OpenDatasets/PKU-SCV