Is Axis Right About H.264 Weakness?

Check out this question from an Axis online knowledge test:

Axis says the correct answer, i.e., the weakness, is that 'it is difficult to estimate bandwidth and storage needs.'

The key element here is how much bandwidth and storage varies over time, across scenes, etc. Even for H.264, if you use CBR, than bandwidth and storage is simple to estimate, as bandwidth constant. Of course, Axis does not support CBR. What they list as constant bit rate is essentially VBR with a cap. For background, see: CBR vs VBR: Surveillance Streaming

Many factors impact bandwidth consumption, something we examine in depth in our Bandwidth Guide For Video Surveillance and Advanced Camera Bandwidth Test Results.

Axis does not say what this is a weakness compared to but I am assuming that it is relative to MJPEG (the reference to 'efficient compression standards'). Even then, it is weird to see it at as weakness, since that's like buying a camera from one vendor who says the price is $20 to $100 versus another vendor who says its $500 flat.

Thoughts?


I guess in a sense it's a weakness because with CBR or VBR/w'cap, you can estimate the maximum storage you might need, but the end result might be that you needed less that you thought and bought more storage than you needed.

So it is a weakness of H.264 itself or a weakness of a certain implementation of H.264? And it is a weakness compared to MJPEG?

Some people including engineers prefer situations that are precisely deterministic, even at the expense of being "worse". For those folks, an improvement that reduces certainty is a defict. That psychology may be at work here.

I have not ever seen a switch that will not handle H.264, althought this is an anecdote and others may have a different experience.

Why would losing an I-frame be a big problem? There is always a "next" Iframe to restart with.

My biggest issue the difficulty of stepping backwards frame by frame during forensic review. It can be done but it's quite computation intensive. Motion-JPEG makes this much easier.

In general, any temporal compression introduces a new class of artifacts and H.264 does not escape this problem. That said, most cameras have several dimensions of control of the encoding process and, if the default settings are not acceptable, some compromise can be found.

"Temporal Compression" can you explain that? I know what temporal means but I don't understand your context.

What are the new class of artifacts for H.264? You lost me on this one as well.

John we have a full tutorial on intra-frame and inter-frame (aka temporal) compression here: Surveillance CODEC Guide.

Sure, John. Standard compression like Jpeg takes place seperately on each frame and there is no dependency on preceding frames - that is, no time component.

H.264 (and others) add a second major compression method where frames sent are dependent on preceding (an even future) frames. This is a time based dependency since these frames are a time based sequence. Thus, the "temporal" terminology.

This generally means that only changes in content between frames need to be transmitted. The display tool "adds up" these changes to deliver a complete and updated image for each frame. Because errors happen, all real systems deliver a complete frame (called an iFrame) every so often to serve as a clean restart in case some of these changes got lost.

A full discussion of this topic is beyond my teleprompter script :) but hopefully this helps.

Further, Jpg compression is optimized for continuous tone images. A fairly well-know artifact (for example) of Jpg is it a sort of white blurring on any sharp edges depicted in the image. This is also present in H.264.

H.264 deals with changes across successive frames with additional compression based on motion prediction and other statistical methods. This tends to add motion blur artifacts that are not present in a Motion-JPG stream captured with a fast shutter.

"This tends to add motion blur artifacts that are not present in a Motion-JPG stream captured with a fast shutter."

Concrete examples with images/video please. I ask because we don't see them and we've yet to see someone make a convincing case for such artifacts.