Can you compress MPEG-2 files with h.264 without losing any quality?

[OddTheodora]

Hi,
I have tried to reduce the size of MPEG-2 and MJPEG (Motion JPEG) files by converting them into MP4s. However, either the quality suffers or, retaining the original quality to about 99%, the video size ends up being greater than the source file. So is it actually feasible to use h.264 as a way to save on diskspace? Only with extremely slow processing, or is it some video conversion law that you actually would have to have a source file with better quality in order to attain a smaller file with nearly same quality, because the last percents in quality cost the most processing power and diskspace?
Thanks.

[Brad]

The answer to your thread title is no, but your post is asking a different question. x264 should be able to lower the filesize without significant differences visible when viewing the videos in motion at a given number of screen-widths away, especially if your MJPEGs are high bitrate.

But you haven't told us your conversion settings or the bitrates of your sources.

[OddTheodora]

Yeah, I should have said "almost" any quality loss. Alright, I had an MPEG-2 video at a resolution of 720x596. I used Handbrake for the conversion to h.264. x264 preset was "Slow", so relatively efficient although not maximally efficient. Other setting was "Constant quality" at 10 RF. Preset was 20, recommended for SD material was 18-20. The difference in quality was still noticeable, though, that's why I chose 10 RF. I started the conversion, and the eventual destination file would have ended up with more than 5 GB (which was the source file's size). I stopped the conversion at 30%.

[Kerry56]

Try CRF 16 and Very Slow. That should be overkill for a DVD source.

[Brad]

There are massive cost-benefit jumps between each CRF number. 10 isn't twice as transparent as 20; it's more like a logarithmic scale (or may actually be one since I don't know the exact details). I often use decimal points in my CRF numbers.

Try 17, 16, 15...

[OddTheodora]

Okay, thanks, I'll try your tips out.

[_Al_]

Is that mpeg2 source interlace? Oddly enough, making it double frame rate from interlace you can save some bitrate also, 500kbps?

Your questions is all relative though.
Imagine you had mpeg2 1000kbps. Making H.264 1000kbps will in no way be almost the same as original.
If you have mpeg2 8500kbps, resulted H.264 bitrate might be lower to original bitrate and you might not see any difference.

[OddTheodora]

It's interlaced. By the way, I remember having tried out 15 RF before and noticed that the film grain was greatly reduced (which was a clear sign of quality loss for me).

[poisondeathray]

grain is very difficult to compress

--tune grain might be an appropriate setting to use in your case (it will help preserve grain , but still filesizes will be large)

[_Al_]

https://forum.videohelp.com/threads/344451-the-meaning-of-various-tunings-in-x264

definitively try tuning film or grain, --tune grain or --tune film I use it all the time to get HQ copies, but sure bitrate will go only higher, but as I said you cannot really compare size of original and result if you go for "certain" quality

[OddTheodora]

Wow, this is a science in itself, I'm only beginning to grasp this... Interesting.

[ndjamena]

You could try to 'improve' the watchable quality with some filters, which should help with compression. I'm not sure how Handbrakes Bobbing compares to AVISynths QTGMC de-interlacing but you could give it a try and see if you like it. Then again, last time I checked Handbrake didn't encode as interlaced easily, so you must have used some kind of de-interlacing in the first place. If you're really worried about altering the picture, make sure you don't use loose anamorphic as it blurs the image during resize, strict is the way to go if you crop at all. Personally, I rarely encode anything without some degree of de-noising and I don't see the point of retaining detail if it means retaining noise as well. I find a clean picture with slightly less detail far more watchable than a noisy one, but maybe that's just me. I've switched to CRF14 and can still get files down to half the size after applying QTGMC and undoing cropping/removing anamorphic with NNEDI3_Resize16. The final product looks better than the original ever did, however, QTGMC alters every pixel in an image meaning the original picture is gone already, I wouldn't resize anything that was originally progressive.

[Hoser Rob]

... Personally, I rarely encode anything without some degree of de-noising and I don't see the point of retaining detail if it means retaining noise as well. I find a clean picture with slightly less detail far more watchable than a noisy one, but maybe that's just me ...

I find that too, but the thing with quality is it's kind of subjective. For me it's sort of analogous to audio quality ... I've been a bit of a hifi nut for ages.

With audio in theory you're not supposed to add anything to or subtract anything from the signal.

In practice, if there are some things missing, your brain will fill in the parts that aren't there. Eg. very few people actually are hearing much low bass at all. It's all second harmonic. But your ear fills in the fundamental so you think you're hearing it.

However, if there are things added that aren't in the original recording, your ear can't do this. Added audio distortion ... like noise/grain in video ... are much more annoying.

I actually usually don't use filters that much in the encode process. I usually do it in playback ... smplayer had really good, intuitive filters. Maybe I'm just lazy.

Paradoxically, I find adding noise (dithering) very effective with grainy source videos.