A lot has been written on the relative merits of these processes and how good they are compared to the lowly vcd at a data rate of 1150kb/s.

I decided to do some tests using the same source file. The source material is DV video PAL 720 x 576 as my D8 camera generates. It was carefully chosen to provide 30 seconds of edited vision containing a slow zoom out and a slow pan of the Honolulu skyline which gives plenty of scope to generate wobbly verticals in the pan and truly test the process.

It was deliberately edited to have a main title, 3 sub titles and transitions that will really test the encoders performance. I used Panasonic for Mpeg-1 and LSX3 for all Mpeg2 files. The playback was via the Hollywood Plus decoder card on the Computer screen and on a Studio quality professional 14" TV monitor in svideo and in composite at the full data rate and 25fps from my 233 mmx computer and Pioneer 104 DVDrom - oh yes I burnt a Cd and did not just test from the hdd, no difference anyway.

The results were transferred to VHS tape for comparison as well. This has taken days to complete.

The base avi file for this takes up 216M per minute of vision and the edited render takes 10 minutes per minute of video. The intermediate avi to produce a 352 x 288 vcd size frame takes 40 minutes per minute of video with 1 keyframe in every 5. The reason for that is the DV codec does not allow resizing and so I cannot produce a 704 x 576 resize from the camera source or a 352 x 288 either. The quickest method of producing a VCD for me is 13.5 minutes per minute of video but that process gives wobbly verticals in a slow skyline pan and I have no intention of dropping slow pans.

9 files were made to compare the results, the best images were used for comparison purposes from each group.

Let me say from the outset NO test in Mpeg-1 or 2 equalled the edited DV image quality, directly to the TV monitor or from tape. Consumer encoders?????

1) Mpeg2 based - Using the LSX3 encoder I encoded a variable bit rate file DVD compliant at the full frame size of 720 x 576 with a peak data rate of 9.8Mb/s (this process is said to produce high resolution detail with a smaller file size). The process takes 2.67 hours per minute of video and created a file of 43.4M/minute of video. Its not a usable method in my view to produce even 20 minutes of video.

2) SVCD's are based on Mpeg2 and I am not exactly sure of the specifications so I generated 3 files from a 352 x 288 frame source which took on average 63 minutes per minute of video. I used VBR peak 9.8Mb/s (30.5M), CBR 9.8Mb/s (78M) and CBR 3.5Mb/s (30.4M). I believe these data rates are in excess of what is used in SVCD. All of these files gave images of a lower quality than the one generated at the full frame size. To my amazement, however, there was very little difference in the 3 files when compared individually against each other and the average person was not going to see the difference.

3) MPEG-1. It is said that high data rate vcd's are noticably better so I generated two files at 3000kB/s using 23.6M of file space and the fatest was 14.5 minutes per minute of video and the best quality took 60 minutes per minute of video. This last file was of better quality than any of the Mpeg2 files based on the same frame size.

4) My normal best VCD standard data rate files at 5.04M per minute of disk space and 40 minutes per minute of video to create was used so that no wobbly verticals occurred.

I am not lucky enough to have a DVD player that playes anything more than my normal vcd's but the decoder card allowed a direct comparison.

CONCLUSION:

Just because its Mpeg2 based does not mean its the best. I produced a Mpeg-1 file just a shade better than the best of the small frame mpeg2's. Specifications of the mpeg2 files used I would estimate were at higher data rates than used for SVCD.

The high data rate Mpeg-1 files in Svideo were just better than my normal VCD file at almost a third of the data rate.

Viewed with a composite source I was hard pushed to see a difference and a friend sat in front the of monitor and chose the normal VCD version from three viewings of all files.

So I cannot help but think that if you are seeing a vast difference between your ordinary VCD and the high resolution version then the process to make it can be improved. If you see an MPeg2 based file as better then there is something quite wrong somewhere.

A lot of posters have said that this or that is incredible better, in answer to that:

I know quality is subjective, BUT HAVE YOU ACTUALLY DONE A TRUE COMPARISON TEST FROM THE SAME SOURCE MATERIAL? I think you should before you actually put pen to post!

-- Ross McL (rmclennan@esc.net.au), February 27, 2000

Answers

hey Ross,

im not to sure exact;y who thats directed too ,but as far as im concerned the higher rate (for me 2500) vcds do look a lot better then the standard 115. I capture at a rate of 30000 and here i am only going to 2500 which is extremely close ot what i capture at. my whole thing now with the super vcds i might not bother with and i will tell you why. I capture at 3000 then convert to 2500 so wither way if its a highrate vce or svcd my product is going to look the same. Becasue i am using the same end file for both, so i guess yeah its cool that i can do it, but ill probably just stick with the higher end vcd. The i-author program will let me make menus and FINALLY put stills on a vcd, etc. The program will let me do this for both vcd and svcd--- oh i completely forgot, so yes i do compare what it looks like one way compared to the other, especially if thier is a lot of fast motion the high rate vcd handels this a lot better. Also its good if your source doesnt look the best to begin with. With this higher rate i had absolutely no kind of intereference at all.

-- Doug (mazinz@aol.com), February 27, 2000.

Hi Doug

The interesting thing is that you capture in mpeg and process from that source and are deciding to use the HR VCD because it looks a LOT better.

I capture in full DV and use NLE techniques and the difference is just noticable but takes a lot longer to achieve I bet. I also do not have a set top DVD player that plays HR VCD's and that means I want to develope a system to get the best I can from a standard VCD. Funny enough I have not yet made a full VCD using the techniques I used for this test.

Tell ya what, its a great learning curve, without it I would still think I had the best I can produce and this latest is now 4 quality levels up on what I originally produced!

Oh, it was not aimed, as you put it, at anyone, just a statement of fact based on testing which just happened to be at the same time as you were posting your observations.

Your results and mine indicate to me that vcd's based on the mpeg source seem to benifit from the increased data rate more than the NLE avi based vcd production. This sort of begs the question how different are the two methods in the quality stakes? Is the NLE technique producing a better product? Interesting questions!

Actually I do not want to be restricted to one type of set top to play VCD's, every customer in the world does not have a Pioneer 525, maybe they and we should.

Actually Doug I got the best small frame picture from a Mpeg-1 at 4000kb/s, but I do not want to carry my computer into the lounge each time so as to use the decoder card, it sure gives a set top DVD player a fright in the quality stakes. Seems to me that RealMagic should put a Pioneer DVD 104 and the Hollywood Pro card into a black box and blow them all away.

-- Ross McL (rmclennan@esc.net.au), February 27, 2000.

Ross

I refer to

2) SVCD's are based on Mpeg2 and I am not exactly sure of the specifications so I generated 3 files from a 352 x 288 frame source ...

To what I think, the picture quality of vcd, to a very large extent, is constrained by its limited resolution, ie. 352 x 288. Both SVCD and DVD are developed to overcome such limitation. Using 352 x 288 source to create SVCD isn't theoretically appropriate to me. Since SVCD is designed to handle 576 x 480, a good avi source will be better than 352x288 source. This is at least enable SVCD to capture as much detail as possible in the conversion process. In this way, a sharper image will be produced.

Higher bitrates wouldn't help much if the source frames do not contain enough details in the first place.

-- Daniel Lee (siangneng@sp.edu.sg), February 28, 2000.

Thanks Daniel

Actually I have worked that out from reading todays postings and I will of course do something about it and do a re-check shortly. Quite surprised others did not blow me out of the water - its all a great learning curve even though I have not got much interest in it as a real possibility because my DVD player will not play anything other than a standard VCD and a lot of customers are in the same boat. But for interest I will test it on the Pioneer 104/RealMagic decoder that has given me some excellent results and a method of improving by comparison and does what the 525 DVD does it seems.

What interests me now is the differences that seem to exist between those using mpeg captures as the base and those of us using NLE based DV avi codecs to develope source material. Despite my error in SVCD the HR VCD does not show a big difference on my equipment until I got to a data rate over 4000kbits/s. I am just trying 6000 in mpeg2 from a timeline encoder for Premiere to see if I can get equivalent results in less time by cutting out the intermediate stage.

Thanks once again for the "soft" blow out - appreciated!

-- Ross McL (rmclennan@esc.net.au), February 28, 2000.

Ross & Daniel, My ATI AIW shows better quality at 4200000b/sec for Mpeg2 and anything above that will generates artifacts in the scan lines on the frame. Mpeg1 sweet spot resides at 1983200b/sec and the same goes as mpeg2 above. As for details as Daniel has mentioned, NTSC D1 720x480 does have more details at these given rates without much increase in file size. I am leaning more toward this frame size now than bitrates or SVCD to create XVCD format for my use, since my DVD player can handle this frame size and bitrates without any problem. I'll try to decrease the bitrates and see if resolution and details will demize or not. I hope this will work which means more play time per CD with the same resolution.

-- lnguyen (wingstarzz@hotmail.com), February 28, 2000.

Long

Thanks for the comments - its interesting that the figure you quote for Mpeg-1 as the point of diminishing returns is exactly what the DVD compliant Mpeg-1 SIF Combo is based on and it plays very well on my computer but I cannot check it in the DVD player, I do not have a 525.

There is a lot more in all of this than meets the eye!

-- Ross McL (rmclennan@esc.net.au), February 29, 2000.