Compro VideoMate Ultra
∞(click for 20 frames of HuffYUV, ~8MB)
It's not perfect, but good enough to show flaws in the source -- the broadcaster's source, that is. Those streaks on the bottom aren't analog artifacts, they're rendering sloppiness in the studio's computer overlay. Click the review link above if you don't believe me.
Hauppauge WinTV-401, c. 2002
That's all well & good, but hardly magical until you realize what the other cards think the signal looks like. Hauppauge's 401 was a longtime crowd pleaser for its working VfW drivers under NT/2K/XP, but its tuner just doesn't stand a chance.
∞(~4.5MB)
The black levels are clearly clipping. Sorry -- this is a test of cards' performance out of the box, and NTSC BT8x8s are notorious for needing tweakage in this department. It's trivial enough to punch up the gamma until it looks more like the Compro's default, but you can't create signal from nothing. Oh well.
Side note for CS junkies: dig how dramatically better Huff is able to compress this clip. I'll let you Google for the underlying mechanics of Huffman trees yourself, but suffice to say the algorithm likes you to use as few "letters" of the "alphabet" as possible. Typical text only uses about a quarter of the 8-bit ASCII table, so it compresses very well indeed (much better than 25%, even, because some letters occur far more frequently than others). What does that have to do with video? Well, it's a simple isomorphism -- Y, U, and V values are letters, and the 8-bit gamut is the alphabet. Actually, the computer doesn't even care about the conceptual difference since BenRG's implementation uses a static, predetermined tree (as any realtime codec must). I'll bet if you used this clip to build a new tree, the compression would be even better.
Why? Oh yes, because of my original point: when your video is clipping like this, chances are also failing to saturate the white point. As it turns out, if you look at it with a histogram tool, you'll see that a large chunk of the "alphabet" is going unused. So it goes.
Why? Oh yes, because of my original point: when your video is clipping like this, chances are also failing to saturate the white point. As it turns out, if you look at it with a histogram tool, you'll see that a large chunk of the "alphabet" is going unused. So it goes.
Even if you look past the dull colors for a second, there is still no comparison. High frequencies are totally unaccounted for, and the background noise is much more visible. (Step through the vid if you're not attuned to distinguishing mild analog snow from the CG mountains in a still shot). On the plus side, there's not too much comb artifacting. I'll note here that this test was also designed to determine the effect of poor coax cabling on reception. Raws exist for each card hooked to both my best and worst cable setup, but thus it hasn't mattered. Lossless Huff for the two Compro caps are within 12KB (0.15%); on the Hauppauge, the filesizes are exactly identical. [TODO: objective comparison via PeachSmoother's noise metering.] More importantly, they look identical. Trust me.
STB TV-PCI, c. 1998
The old STB I found in my closet is another story. Dot crawl and rainbows were apparently a fact of life back then:
∞(~5.1MB)
Not terrible, and no less detail than its newer BT878 cousins. Hook the tuner up to a source that's sensitive to RFI, though, and the comb effects really blossom:
∞(~5.6MB)
Yup, that's a 10% increase in filesize (read: additional colors added to the "alphabet") just from rainbows. Moral: if you must buy an ancient cheapo card, at least spend a few bucks on cabling.
Hauppauge WinTV-PVR150MCE
That was then, and this is now. With what kind of tuning hardware is Hauppauge equipping their latest offerings? The PVR-150 is only a few months old, representing their third-generation foray into consumer mpeg2 cards. Let's have a look:
∞(~6.1MB)
Clearly a step or two above the BT cards, though not the Compro's equal. Detail isn't resolved quite as well, as you can see when picking out the mountains (both the good ones and the misrendered ones) against a shadow. It doesn't help that levels are more comparable to the clipping-fest given by stock BT cards (note Huff's filesize agreement); not much you can do in post, and worse, I don't know of any PVR programs (for which such a card will certainly be used) that let you tweak the input amps or luma coring or anything of the sort. Finally, there is moderate comb artifacting -- in the still you can see that solid colors near hard edges aren't uniformly saturated, and sure enough if you download the video clip you'll see dots crawl. This is nothing AviSynth can't cure, however.
If you're wondering about the mpeg2 compression, click to the full review. I don't think it's a factor in our comparison, but FWIW it's set here to max quality (about 12Mbps), then recompressed into lossless Huff like everything else.
Back to VidCapSamples
There are no comments on this page. [Add comment]