Elevating Video Quality With the Video Compression Score Metric

In today’s media landscape, ensuring the delivery of high-quality content has become a top priority for service providers. Not only does it impact the viewing experience for an audience that is more discerning than ever, but it’s directly tied to customer satisfaction and retention, and therefore the bottom line. Compounding the issue is the ever-expanding volume of content being delivered. To effectively manage this enormous data flow, video content must be compressed before transmission, a process that can result in the degradation of content quality. The level of degradation depends on the temporal and spatial complexity of the content and encoding methods being adopted by the transcoders. With low-motion content like news programs, there are very small changes within frames (spatial domain) and across frames (temporal domain) that only require minor adjustments in encoding.

With high-motion or high-textured content like action movies or car races, there are major changes in the spatial and temporal domains, which necessitate an increase in the bits required for encoding. To handle this increased bandwidth requirement, the video encoder must either supply the number of bits the content demands or be adaptive enough to change the encoding method. If not handled correctly, artifacts can occur that impact the viewing experience, such as blockiness, blurriness, flickering, and more. To estimate compression degradation — a metric known as the video compression score — reference-based or non-reference-based methods can be used. The reference-based approach involves comparing the compressed video with the original content, while the non-reference-based approach doesn’t require this comparison. 

This paper will focus on a non-reference-based approach for calculating the video compression score that utilizes the encoded video parameters of the compression bitstream. Attendees will discover the advantages of this approach for a variety of scenarios. These will include real-time quality monitoring for live video streaming, video conferencing, or security applications in which a reference video is often not available, and providing an objective quality assessment for automation, quality control, and troubleshooting purposes. To achieve the best results, they will learn how an AI neural network can be trained to estimate the video compression score; categorize the transcoded video as unacceptable, marginal, acceptable, and excellent; and correlate the results with well-known video quality methods such as Netflix’s VMAF. The result is a highly accurate, efficient, and cost-effective metric that can be used in real-time applications across the IPTV, OTT, and post-production markets.

Shekhar Madnani | Interra Systems | Cupertino, Calif., United States
Yogender Singh Choudhary | Interra Systems | Cupertino, Calif., United States
Muneesh Sharma | Interra Systems | Cupertino, Calif., United States