HandBrake

In a previous discussion, we compared Bob and Decomb from some noisy SD analog footage, and I came to the conclusion that the bobbed footage was visibly noisier and less detailed in the shadows. For the purpose of the discussion, "Bob" means 60p output, and "Decomb" is 30p, using the defaults at RF20.

Using a Belle-Nuit 1080 test chart, I rendered Sony YUV 8-bit 4:2:2 intermediates and handed them to Handbrake. As expected, the chroma subsampling flags triggered from the downsampling to 4:2:0, so what we're interested in is only between 0-16 RGB below. Then we compared them to Project previews at 60p (59.94) and 60i (29.97), using a Difference Mask to isolate noise content.

Both fared well with static images. Decomb was sharper and cleaner on the scopes, but I really couldn't see a difference in the delivery. Bob had more shadow noise, but it was also difficult to detect in a static scene (still image). Here's the decomb 30p noise print.


And here's Bob with a static image.


I predicted that noise would be worse with motion detail, and this was reflected in the tests. Again, Bob fared worse. Here's the Decomb motion noise print.


And here's Bob under motion. This time, I could see a slight difference during fades, which are particularly bad for showing high noise vs. detail.

My Preliminary Conclusions:
Both Bob and Decomb give excellent results with HD.
The areas where Bob seems to break down sooner are motion detail, shadow detail, fades, and noisy or grainy source, especially analog source.

The things one can do when dealing with analog source with either deinterlacing approach are:
-- Run it through NeatVideo before Handbrake (untried)
-- Chop off 0-8 RGB in an editor before Handbrake (mostly noise and practically no information there).
-- [Deleted]
-- When dealing with analog gameplay, consider encoding 30p instead.

musicvid wrote:-- Raise the vbv-minrate.

Where? x264 has no such option

Right you are, I think what we did was lower the RF but leave vbv-maxrate capped to get more bits in the shadows. Larger files, of course.
Another test to come, to see if this works.

Musicvid, I have a question for you. What is shadow noise? I know you mentioned it in our last forum but couldn't see what you were talking about.
I do expect bobbing 480i to show a lower quality still image as its output is essentially 720x240 output. 1080i is 1920x540 bobbed

It's what you see in the two frames I posted from your source in the previous thread. I'm certain you see the difference.

what am i looking for? I couldn't see what you are talking about. I'm assuming its the mpeg4 gradients in shadows due to too low of a bit rate, but could be wrong.

One image is sharper.

I don't know if handbrakes bob filter does yadif correction like the default decomb method. (could any handbrake developer clear this up?) If not, the single frame results would defiantly be sharper on the 30P video. Other programs can do bob with yadif, or other words ,bob + weave deinterlacing. Also my 60P video, I didn't account for the higher frame rate, so less bits to each frame, hence less sharp of an image. Can't compare the 2 videos for quality differnces.

musicvid wrote:When dealing with analog gameplay, consider encoding 30p instead.

Why do that when the original gameplay is at 60FPS?
I would encode movies at 24FPS, or if needed detelecine them to 24FPS
Most sitcoms run at 30FPS, and I would encode those at 30FPS. No bobbing or deinterlacing, even if the source is 1080i.
Every frame rate has its place and 1080i (although not ideal) is just a transportation method that can carry all 3 frame rates. I just like to preserve them.

I suggested you guys start your own thread for game capture discussion; now would be the perfect time to do it.