A GPGPU-based software implementation of the PBDI deinterlacing algorithm
Gilbert Kowarzyk, Normand Bélanger, Yvon Savaria · 2011
We present techniques for creating a software-only high-performance implementation of the Pattern-Based Directional Interpolation (PBDI) intra-field deinterlacing algorithm. The proposed techniques use General-Purpose computing on GPUs (GPGPU) for accelerating the computation of missing pixels in the interlaced fields, focusing on maximizing the available bandwidth and parallel computing resources offered by the GPU. This work demonstrates that ASIP like performance can be obtained with GPUs with a much lower development effort measured in weeks instead of months. The reported GPU implementation almost matches the performance of a fully optimized ASIP obtained after several months of research effort. Furthermore, this flexible software-only technique allows for easily updating the end-user systems as improvements in the algorithm are made.