Click Here for Video

2009 Student HPC Competition Title: No-reference Video Quality Asssessment on HPC with interaction of MPI C++ and Matlab By: Ji Shen

Abstract

In this paper, we propose a new Video Quality Assessment (VQA) method based on curvelet and cosine transforms, called Video Hybrid No-reference (VHNR) method. VQNR is capable of predicting the quality of noisy, blurred, or MPEG2 compressed videos and requires no high quality reference video. The model is based on obtaining a statistical relation between the video quality and some video characteristics. Therefore, we must analyze tens of thousands of simulated videos. A video pool is continuously lled by several instances of Matlab programs which build the simulated videos. At the same time, the video is fetched by a C++ MPI program which does 3D curvelet transform using 16 nodes on HPC. The histogram of curvelet coecients are saved for further matlab processing. The intensive computation requirements of 3D curvelet transform and processing of tens of thousands of videos require the power of HPC.