Testing HCenc against pay mpeg-2 encoders (PSNR and visual)

[jsquare]

Well to answer my own question I decided to run my own tests and here is the setup:

Encoded 3 different titles using DVD-RB with encoder default matrix and no filters, movie only, first audio track, no subs, re-authored w/o compression using Shrink.

The encoders, HC v0.21, CCE SP2, Procoder 3 and DVDShrink as reference.

MSU for PSNR metrics and 37" LCD with JVC progressive DVD player via component for visual comparison.

PSNR Tests
First title - Stargate, Ultimate Edition 119mins:
1.DVDShrink
2.HC 2-Pass
3.HC OPV
4.Procoder
5.CCE 2-Pass

Second title - Transformers 143mins
1.HC 2-Pass
2.Procoder
3.DVDShrink
4.CCE 2-Pass

Third title - Lawrence of Arabia, Superbit 227mins (2-disc reauthored into one)
1.HC 2-Pass
2.Procoder
3.CCE 2-Pass
4.DVDShrink (Not tested, title too big, must be done in 2-Pass)

Visual Tests
Titles 1 and 2
1.DVDShrink
2.Procoder
3.HC
4.CCE

Title 3
1.HC
2.Procoder
3.CCE
4.DVDShrink(Not tested since I know that it will look so bad with 2-pass trans-coding)

Fastest on average
1.DVDShrink
2.CCE
4.HC
5.Procoder

HC is clearly the winner, while Procoder still holds some value on the visual department, but CCE was a complete disappointment after all these years and different revision the encoder has not improved.

So the answer is simple, Shrink for 70% or higher compression and HC for the rest.

[kolak]

CCE is an encoder to use with a proper uncompressed source (straight From Beta tape) and preferably film based. You're using Mpeg2 as a source and this is a different story.

Try compare CCE and other with nice film based uncompressed source and you will see how nicelly CCE preserves details (eg. ProCoder smooths them a lot).

Cinemacraft is working now to improve their HD encoder and they don't spent so much time to improve SD encoders.
I would like them to optimized CCE for multicore processors to get 20x faster than realtime encoding

[Irakli]

Interesting test, jsquare. I thought that encoders are always better than transcoders regardless compression ratio, but this test shows that everything is not that obvious.

Regards,
Irakli.

P.S.: I encode my home videos from DV format to DVD (without deinterlacing, i.e. output DVD stream is interlaced). Could anyone suggest me which encoder should I use to preserve maximum details at bitrates over 8000kbps? (I refer to free encoders like HC, Quenc and ffmpeg; the only non-free encoder I own is TMPGenc)
Thanks.

[prOnorama]

Originally Posted by jsquare

MSU for PSNR metrics and 37" LCD with JVC progressive DVD player via component for visual comparison.

A 37" LCD? Maybe try a 50+" Pioneer/Panasonic plasma, the difference between transcode/re-encode might not be noticable on smaller screens

Originally Posted by jsquare

So the answer is simple, Shrink for 70% or higher compression and HC for the rest.

I don't think it's that simple. Personally I would not use DVDShrink for < 85% compression.

Here's a blind test done some time ago:

Not what I expected, kinda at a lose of words, but clearly DVD-RB w/ CCE Basic looked better then both DVDShrink and IC8 to the 37 people I showed them to.

This test was just the main movie and one audio track ranging from around 90% to 70% compression.

http://forum.doom9.org/showthread.php?t=91418

DVDShrink has not improved since then -development discontinued in 2004- but encoders like HC, CCE & Procoder have....

[Boulder]

Originally Posted by Irakli

P.S.: I encode my home videos from DV format to DVD (without deinterlacing, i.e. output DVD stream is interlaced). Could anyone suggest me which encoder should I use to preserve maximum details at bitrates over 8000kbps? (I refer to free encoders like HC, Quenc and ffmpeg; the only non-free encoder I own is TMPGenc)
Thanks.

I'd try using QuEnc at constant quant mode at q2 with interlaced encoding. It's probably slower than HC but I think dragongodz has made some tweaks regarding interlaced encoding.

[Irakli]

@Boulder

[tahir]

Originally Posted by Boulder

I'd try using QuEnc at constant quant mode at q2 with interlaced encoding. It's probably slower than HC but I think dragongodz has made some tweaks regarding interlaced encoding.

AFIK

For High Speed = CCE
For High Quality = QuENC

this was about a year ago, i havent used Henc

[davidk21770]

Does anyone have a feel for how TMPGEnc fits in here?
Probably a bit slower, but quality?

[jsquare]

This week I went back to my old Sharp 55" RPTV, the main reason is that I wanted to test my new Toshiba HD-A3 HD-DVD player on a bigger screen than my 37" LCD, since I didn't noticed much of a difference between SD-DVD and HD-DVD(maybe a 10% improvement). To be honest size does matter when it comes to HD, HD-DVDs do look better on my Sharp even at 1080i via component vs 1080i via HDMI on my 37" LCD.

But this time Shrink won on visual tests for titles 1&2. The Transformers encode made by HC/Procdoder/CCE showed some serious blocks on semi-static scenes and dark areas, while Shrink looked some much like the original. Some people say that the HD-A3 is one of the best 1080i up-converter, but until now I only experienced with with my XBOX(XBMC) and with my HTPC. The HD-A3 seems like a very discriminative player and it really shows the flaws on the encoded material.

Once again I wanted to try something different so I tried a very bad source, the TV series LEXX S2V1 4-epidodes disc, which are very noisy and badly transferred/authored DVDs. This time didn't bother with PSNR just visual testing and to be honest Shrink came out ahead once again, followed on a tied for 2nd place by HC/Procoder and at the bottom CCE.

[avdw]

Can anyone tell me, for best quality, which one is preferable ?

HcEnc or QuEnc ?

[urvieh]

I'd say the HC, especially the latest version with smp capability.

[avdw]

Originally Posted by urvieh

I'd say the HC, especially the latest version with smp capability.

Thanks a lot urvieh

[Pulp Catalyst]

jsquare,

i ran some of my own tests, and from what i can tell, if the source is good, then shrink doesn't stand a chance, it smooths the image to much,

are you sure your zooming in a a still frame captured on a finished encode,

as like jpeg this is a good way of testing, and shrink, like many transcoders cheats, where as it hides the blocks by smoothing them out, but this causes massive amount of detail lost,

most transcoders does this to hide the blocks, where as encoders are not allowed to do this, as encoders have to rebuild the frames from ground up, so the detail is preserved and no smoothing should ever be used, (if a user wanted the material to be smoothed then a filter could easily be used)

procoder has the edge just, but HCenc is right there with it,

CCE comes in last place but only for low bit rates, anything over 2500+ or on complex material 3000+ then CCE wins hands down,

but for the record, CCE was never designed for low bitrates, CCE was designed for hollywood standards encoding (in other words, high bit rate conversions destined for DVD-9 disks),

CCE is great for just film, 3 passes, but if the film is complexed and over 2 hours long then forget it, CCE can not handle this type of compression on a DVD-5 disk against the other encoder mentioned above,

my money is on HCenc as it's already level pegging with procoder and i feel it has a better future, for low bit rates, but only just, but i'm sure this is debatable, due to other variables to consider (filters, and more importantly matrices and type of material being encoded)

and to my knowledge, a transcoder can never be as good as a encoder, a encoder rebuild the frames from scratch, a find it hard to believe a transcoder can beet this concept, giving the nature how a transcoder works,

the one thing that does surprise me though is speed, HCEnc is excellent, but it's very slow, i don't understand why cinema craft don't get there encoder to take more time on each frame, and try to get it up there with the others, as CCE is the fastest encoder around (that still brings so much quality to the stage)

if they only stuck with there very expensive flagship product then no need,
but they didn't, as soon as they released there CCE basic program, then this immediately put them in concession, and they owe it to there customers to give the quality they deserve, and to me it seems like they have abandoned there SD product (or perhaps they felt it was time to get out......quick) either way, i feel this is wrong,

no disrespect intended jsquare, i have no intentions on stepping on toes, just thought i would put my 100 2 pennies in,

respect,

[Wodan]

doesnt the 70% shrink rule kind of depend on the movie? if its a movie thats kind of grainy in the first place...wouldnt it be better to use an encoder even if the compression was only lets say 80% ?