Honey,I Shrunk the VIDEO!
a&s International | Date:
The H.264 compression standard comprises a variety of profiles, each of which provides a rich set of features. The various profiles are targeted for different applications and, depending on the features enabled in the encoder, image quality and compression efficiency can vary widely. Sometimes, H.264 and the supported profiles are promoted by vendors as an assurance of quality,but this notion is often misleading.
Image quality and compression ratios are generally poorly specified, and there is an emerging realization that there is a big difference between good and bad encoders, even when their specifications on paper look identical, said Bengt Christensson, Senior Marketing Director at Ambarella. “The H.264 specification specifies the function of the decoder, not the encoder. A comparison of the video quality and storage use between two DVRs with identical specs may look very different in reality. The difference between a good and a mainstream H.264 encoder may be up to four times in compression efficiency or image quality, and even more when capturing complex scenes with motion.”
Rather than which H.264 profile is supported, it is more important to consider which specific features of H.264 are enabled. For instance, low latency is important for viewing streams, while support for the best-quality compression is important for storage streams, Christensson said. “An emerging demand for H.264 features we see for viewing streams is baseline or main profile with infinite groups of pictures and intra refresh. This is similar to what video conferencing is using, and it enables lower delay and no I-frame beating. Common features for storage streams are main or high profile with B frames. This is similar to what broadcasting is using and enables best quality and compression ratio.”
In the past, security applications mainly used baseline because it is less demanding on the processor, Golston said. “More recently, solutions have introduced the ability to use high profile. It delivers a significant lift in the amount of compression you can get. Many systems today can easily meet the requirements, so there is increasingly less technical reason to stick to baseline. However, there are tools in high profile that should not be enabled in some security applications, such as those that introduce additional latency.”
“DVR manufacturers are under no obligation to apply all of the tools that H.264 affords, and most vendors only use a small subset of the tools available,” Oliver added. “There can be a huge difference in the encoded video quality or bit rate between devices that claim identical specs.” The focus is not so much on which profiles or tools are used, as a good H.264 baseline encoder can easily outperform a so-so high-profile encoder.
Where the Gap Lies
The algorit hms used for compression create a wide spectrum of variance in performance. “For standard definition (SD), the differences may be less — perhaps two to three times — since mainstream chips have enough computing power to keep up with motion estimation when there are fewer pixels to process. Making a good SD chip is a lot less complex than offering a good HD chip,” Christensson said.
In the past, M-JPEG and MPEG-4 DVRs displayed comparable performance; today's H.264 encoders provide many more possibilities for enhancing video compression, Christensson continued. “HD resolution is at least six times that of SD, and H.264 is roughly four times more compute-intensive than MPEG-4. An HD H.264 chip requires 24 times the processing performance of an SD MPEG-4 chip.”
“In addition to system level trade-offs, such as the ratio of I frames used, the encoder algorithm technology used, including the features used for motion estimation and rate control, create a huge difference in the bit rate and quality you can reach between different implementations in the market — even with a specific codec profile,” Golston said.
“The differences between two DVRs (or network cameras) can be substantial when using H.264. A broadcast-grade H.264 encoder may be able to compress video up to 10 times better than a mediocre H.264 encoder in complex scenes with motion,” Christensson added.
One encoder specification to look for is the motion estimation search range value. “Some low-cost encoders only have about +/– 36 by 36 pixels. More advanced devices use broadcast-grade search ranges, which are all the way up to a full-frame width search range (1,920 pixels),” Christensson said. “The most complex part of an H.264 encoder is the motion estimation, which assists in compression of moving objects. There are encoders with very poor motion estimation that will generate bit rate spikes, which can lead to packet loss, or blocky artifacts at moving objects or when panning/zooming a dome.”
Another important factor to consider is what rate control to use. For networks where a bandwidth budget is specified, constant bit rate can be used, whereas variable bit rate is used for situations where constant quality is important and sufficient network bandwidth headroom is available, Christensson continued. “There is a wide variety in how well rate control is implemented; more advanced network cameras and DVRs support macroblocklevel adjustment of quantization parameter, and allow for many tuning settings such as maximum and minimum quality when a certain bit rate is reached.”