Taking Advantage of Heterogeneous Platforms in Image and Video Processing
Sidi Ahmed Mahmoudi, Erencan Ozkan, Pierre Manneback, Süleyman Tosun · 2014
This chapter proposes a model for image and video processing which exploits parallel (GPU, graphics processing unit) and heterogeneous (multi-central processing unit (CPU)/multi-GPU) architectures. It categorizes and describes two types of related works based on the exploitation of parallel and heterogeneous platforms for multimedia processing: one is related to image processing on GPU, and other is devoted to video processing on GPU. It presents the proposed model for image processing on GPU. Then, it describes the development scheme for processing multiple images on heterogeneous architectures. It also summarizes the proposed scheme for real-time video processing on GPU. The chapter describes two use cases of the presented model with a comparison between CPU, GPU, and hybrid implementations. Finally, it concludes that this model offers an efficient scheduling of jobs based on estimating the performance and the required transfer time and enables overlapping data transfers by kernel executions within multiple GPUs.