Updated: I’ve done a little research on image evaluation, and decided that SSIM presents the best option to evaluate the the perceptual quality. SSIM results are now included below.

In a recent thread on the Avid Community Forums there was some discussion of the quality of Avid’s downconversion from HD to SD. I have used various downconvert methods from within Media Composer in the past and have never had any serious problems with any of the results, but I figured I might be able to conduct some vaguely reasonable tests.

The Methodolgy

I captured 1080i50 footage I had shot on XDCAM HD422 with a Sony PDW-700, from a Sony PDW-HD1500 via HD-SDI into Avid Media Composer 4.0.2 with HD Adrenaline hardware as DNxHD 185X. I then used a variety of methods to Digital Cut the footage back to a Digibeta tape, recaptured the footage from that Digitised that footage back into an PAL project as 1:1 10bit and export uncompressed stills of a frame that showed sufficent detail and compared those frames to one another in Adobe Photoshop.

The Methods

Avid Software – I changed the Format of the project from 1080i50 to PAL 25i and created an SD sequence which I Digital Cut to tape, ignore Avid’s suggestion to transcode first.

Avid Hardware – With the project mode set to 1080i50 I created an HD sequence and set the Video Output settings to enable an Anamorphic Downconvert output. I then Digital Cut this to tape from the SD-SDI output of the Adrenaline.

Transcode – With the project mode in PAL 25i I copied my HD sequence and used the Transcode tool to convert it to 1:1 10bit media, then Digital Cut this to tape as usual.

HDCAM – With the project mode in 1080i50 I routed the HD-SDI output from Avid to a HDW-D1800 HDCAM deck’s HD input, then routed the SD-SDI output from that deck to my Digibeta which I controlled from Avid. I Digital Cut the sequence to the Digi deck with video passing through the HDCAM deck.

HDCAM SR – As with the HDCAM test above, but I routed the video through a SRW-5800 deck.

SD Reference – I set the Avid project mode to PAL 25i and captured the same segment of video from the PDW-HD1500’s SD-SDI output.

Photoshop – I exported a full 1920×1080 still image from the original HD capture and resized to 720×576 in Photoshop using the ‘Bicubic Sharp’ method.

My Observations

To my eye the best SD picture probably came from Photoshop’s resizing – it remained sharp, but that’s probably because it’s not interlace-aware and has created greater sharpness from merging fields.

The best actual SD video seemed to come from the hardware E-to-E conversion through either of the HD decks (only a very very minor gamma difference seemed to be evident), followed very closely by SD output from XDCAM HD deck, but the Avid Hardware downconvert (from the Adrenaline DNxcel hardware) was very very close in quality. Surprisingly there seems to be no appreciable difference between the realtime software and Transcoded output from Avid, which are less sharp than hardware options, but only slightly.

Overall, all the hardware methods (all three decks, and Adrenaline HD realtime downconvert) were incredibly close. With both software-based options being indistinguishable.

To my eye, with repeated plays on a fairly good monitor, I couldn’t see any significant differences between any of the options – all produced images that appeared sharp and correct with no offensive softening or field jitter. I would also speculate that the HD downconversion from newer generation DX hardware will proform at least as well if not better than the older Adrenaline hardware.

The Examples

Below are a number of images which represent a 190×150 section of the sampled frame, blown up 300% (with no blending or softening) – you can compare them to one-another by picking between them below (once you’ve loaded each one once, they’ll toggle instantly)

• Photoshop
• SD Reference
• HDCAM Deck
• HDCAM SR Deck
• Adrenaline Hardware
• Avid Transcoded
• Avid Software

Quality Evaluation
Using the SSIM method I compared the quality of my framegrabs, ordered by quality, as compared to the SD Reference (the hardware SD output from the PDW-HD1500 deck).

1) HDCAM SR Deck – 0.9902
2) HDCAM Deck – 0.9873
3) Avid Adrenaline Hardware – 0.9847
4) Avid Transcode – 0.9293
5) Avid Software Realtime – 0.9290

These results apply to this test only. I will attempt to reproduce the test in the near future and take more samples for a better overview.