Bitrate comparisons

[myst420]

We talk a lot about the effects of bitrate has on the quality of an image. I wanted to test this. After looking all over the web I could not find a source image of sufficient quality to make for a good test. Then I had an idea.

Some games allow you to export screen caps. I recorded about a 5 min loop in Far Cry. I then had it replay and save out each frame of the loop generating about 9500 jpeg's, each about 2mb. This is very low compression and pretty much lossless from the original images. I used this to generate an uncompressed AVI file, ~30gb. I clipped off the uninteresting beginning and end of the clip and the end result was a bit under 3 min (5000 frames)

This source is fairly difficult for a codec because of the high frequency detail textures the game uses. You can see this in the detail of the ground and walls. This is slightly easier to encode than film grain but more difficult than a typical CGI film which is ray traced.

The next step was to do some encoding. Thankfully Ben, the vc-1 expert, had this to say over at AVS. He was also nice enough to provide links and suggested settings for VC-1 which worked very well.

Fully understood. Which is why I keep suggesting that folks download our free encoder and make some samples for themselves to see for themselves .

http://www.avsforum.com/avs-vb/showt...&#post10220450


For AVC I downloaded an utility called megui, which uses the x264 codec for encoding. It took me a few tries to get the settings right to get decent results from this one since I didn't have any good initial suggested settings to start from.

Now on to the results. I compressed the clip several times. I used VC-1 set at 10, 15, and 20Mbps. The public VC-1 encoder is limited to 20Mbps so I could not test higher bitrates. For AVC I tried 15, 20, 25, 30, and 35Mbps. I used a variable bitrate for both allowing the encoder to do its best to distribute the bits. The AVC encoder allowed me to specify a peak rate so I used the 40Mbps peak that BD provides as the cap.

I picked one frame from the original source then used media player classic to capture that frame from each encode. I would suggest downloading the images then using the windows preview feature to flip between them. It makes the differences easier to see.

Each image that follows is an lossless compressed images, about 4MB each.


VC-1 @ 10Mbps
VC-1 @ 15Mbps
VC-1 @ 20Mbps

AVC @ 10 [coming soon]
AVC @ 15Mbps
AVC @ 20Mbps
AVC @ 25Mbps
AVC @ 30Mbps
AVC @ 35 [coming soon]

Finally the reference frame.

I will not tell you what to make of the above images. But I am curious which of the above everyone thinks is “Good enough” and which “They want to watch”