Netflix ingest and encoding pipeline is a cloud-based platform that generates video encodes for the Netflix streaming service. Due to the large throughput of the system, automated video quality assessment of the source videos and the generated encodes is essential in ensuring the quality of experience of Netflix subscribers. Owing to the diversity of video content, traditional video quality assessment methods are not able to meet the need of Netflix video pipeline. A joint research team between USC MCL and Netflix is assembled to tackle this difficult problem.
A scalable solution of video quality assessment method is proposed by Joe Yuchieh Lin (USC), Eddy Chihao Wu (USC), Dr. Ioannis Katsavounidis (Netflix), Dr. Zhi Li (Netflix), Dr. Anne Aaron (Netflix), and Prof. C.-C. Jay Kuo (USC). This method is called Ensemble-Learning-based Video Quality Assessment Index (EVQA). EVQA adopts a frame-based learning mechanism to address the limited training data problem and fuses multiple image quality assessment indices to generate the final video quality score. The superior performance of the proposed EVQA index is demonstrated by experimental results conducted on both LIVE and MCL-V video databases.
The discussion of Netflix’s pipeline system, current solutions and remaining challenges will appear in ICIP 2015. The work of EVQA is accepted by ICME 2015 workshop on Cloud-based Media.