I’ve been spending a lot of time comparing codecs lately, and I always feel a little uncomfortable when I encode the test files since I assume that the actual codec vendors know how to wring the last bit of quality out of their respective codecs better than I do. So for this particular review, I asked the codec vendors themselves to encode the files.
Specifically, I sent a test file to Microsoft (for VC-1) and On2 (for VP6), and each company preprocessed and encoded the clip to a standard set of parameters using their own tools and compression settings. Unfortunately, the company that had agreed to produce the H.264 files backed out. I was unable to find a replacement, so I encoded the files myself.
In the past, for these types of competitions, I sent the participants a prescaled, prefiltered test file so I could isolate codec performance from the qualitative aspects of scaling, deinterlacing, and noise-reduction filtering. For this test, however, I felt that insight into the types of preprocessing performed by these companies to produce their “trophy footage” would provide some interesting perspective into the art of real-world preprocessing and encoding.
So I produced the test file from seven different HDV-source scenes, edited together in Premiere Pro and exported in HDV parameters (1440x1080i) using the Lagarith YUV 12 lossless codec, which was as close as I could come to the original HDV format without imposing another layer of MPEG-2 compression by rendering back to HDV after editing.
The rules were simple. The contestants could use any preprocessing and/or encoding tool available, irrespective of cost, as long as they disclosed the process to me so I could share it with you. Of course, I was free to do the same when I produced the H.264 file.
The streaming files had to be delivered at an average data rate of 800Kbps video/128Kbps audio, which is insanely aggressive, but I felt that it was necessary to distinguish the quality of the three codecs. The companies could use the highest quality settings available (such as VP6E rather than VP6S, or the High Profile rather than Baseline for H.264) and either VBR or CBR, as long as the average data rate came in at about 928Kbps.
Once I had all the files in hand, I compared the quality of each. I am reporting the results herein. Note that you can view the files and all preprocessing and encoding preset files at www.doceo.com/HDcomps.html.