Case study of programmable video post processing: CUDA-based novel edge directed video scaling

Shengqi Yang, Ronghui Cheng, Li Zou

Abstract: This paper explores the possibility of real-time programmable video post processing. As a case study, it implements a novel edge directed video spatial resolution up-conversion algorithm based on the Nvidia CUDA (Compute Unified Device Architect) technology. The algorithm can reliably determine the best direction to perform spatial interpolation. Algorithm and implementation optimizations for execution speedup are discussed according to the properties of the CUDA framework. Experimental results show that this novel algorithm performs better than the industry leading algorithms in terms of both visual quality and computation complexity and is a good candidate for real-time programmable HD video scaling in CUDA.