byuu's message board

Ok, after my headache-inducing observations of my new CRT tv, here's my latest attempt!

Changes:
-New scaling method (faster, less blurry (hopefully!), more faithful to what a real CRT does).
-Correct aspect ratio (the same one bsnes uses anyway, so blame byuu if you don't like it! )
-Smaller phosphors.
-New scanline method.
-Correct number of scanlines (count 'em!)

Caveats: I've only been working on it for a few hours so currently
-It doesn't replicate the scanline saturation described by NFG
-The overall brightness is a little too dark (maybe the scanline roll-off is a bit too strong?)
-It has only been written for 5x scaling (will be changed so it works for all integer scaling!).

Secret of Mana (look at the sparkles in the water! )

Original (left), Scaled (right).

Zelda with blargg's NTSC filtering

Original (left), Scaled (right).

Better? Worse? Let me know what you think!

Thanks!

I like the top one better. Not too blurry.

Yeah that's a lot better, that should match NTSC CRT's pretty nicely.

can't wait to see the results with NFG's style saturation, bright "pixels" should bleed more into adjacent "pixels" and "scanlines" than dark "pixels".

I had that darkness problem too, i think a real CRT is much brighter than the digital image we are working with causing the black+colored lines to tone down to be displayed at the right GAMMA.

I will try to get a screenshot of Silent Hill to work with, because that will be a real good test.(there are also several nes and older console games that depend on CRT stuff to look correct)

yes I like the top one, very good work sir. I like.

wherever those vertical lines on the zelda image are coming from though is quite painful on the eyes. The previous zelda images were far better but the SoM looks great.

for the record I saw no great thing wrong with your original posts, the combination with blarggs filter also looked nice.

and yes I far prefer the scaling and then filtering method, the other way looks like ass.

My web browser (firefox) autoscales the images and causes every few vertical lines to be brighter in the zelda image (it actually occurs in the SoM image too, but it's hardly noticeable). Make sure you're viewing it full-size. (Not saying it's the same issue for you -- but the vertical line problem that concerned me was due to browser scaling and I'm probably not the only one)

Because our displays aren't bright enough to get the same light output when simulating an old TV in this way, and because the resolution isn't high enough to exactly emulate an old TV (with IMO brightness being the bigger issue), it's probably necessary to make some subjective judgments in order to get to the best solution. This happens in art/illustration too. Imagine trying to draw a bright sun or reflection from it on paper -- Even if you leave it plain white, the paper's never as "bright" as you'd probably like.. so you've got to allow for a wide area of saturation (as would occur in a photo for example), make sure there's something around it to contrast with it to provide reference, etc.. and in this, there is even the choice you've got to make about exposure (which in a sense is similar to how the eye has adjusted to brightness beforehand). I guess there's been some mention/hinting at these things in this thread already (and even distance from the screen, which will also vary).. but I find it interesting and it's a pretty deep issue so I figured I'd mention it anyway. It's funny -- If you want a better user experience already, you can recommend that the user keeps the lights off for a while with the curtains drawn, and then turns on just the screen in the dark. A comparison with an 'original' is not necessarily relevant when the user's perception is not a constant (of course this can't work in the daytime and/or with usual surrounding light, but it's something to think about).

Your algorithm already does a better job with brightness and bleed than most that I've seen in the past.. but it is still dark and most people will want something to be done about that. In my opinion, it's good to work on something that seems most technically accurate for a while (mostly ignoring the specifics of the display devices, and their limitations -- almost using a theoretical display device), and then once you've basically mastered that, then you can think about the fudging necessary in consideration of display devices and perception issues without getting everything confused too much. I guess you're basically there now.

If you haven't tried it already, I'd recommend taking a photo of your CRT using a high-res camera and/or allowing the lens to be very slightly out of focus. Then view those images on your target display and see if you're satisfied with that (ignoring any artifacting and whatnot that you couldn't avoid). If the exposure's good and you like how it looks, then you've got some good reference to work with.

As a complication, bear in mind that some displays are brighter than others and have different dpi (ideally, some configurable factors could be tweaked to handle this). My latest has a very bright LED backlight, and a high dpi 1920x1080 resolution on a 15" screen.. so I'm more looking forward to what can be done with CRT simulation than I've been in the past.

I think all that needs to be done now is to simulate the sometimes slightly off geometry experienced in a CRT set, and also the "roundness" of your typical CRT.

Well I can live without that.

DRUMROLL PLEASE

I've managed to get my 5x algorithm fairly close to where I want it to be, and have also managed to shrink it down to 3x and 2x! They're not EXACTLY the same at the moment, but they all work in principle, just need a few tweaks to equalise the gamma etc. (I literally got the 2x algorithm working about half an hour ago!)

Here's a few samples (Make sure to zoom in if your browser automatically scales the image!):


GnG (All sizes)


SoM (Original, 5x, 3x(big phosphors), 3x(small phosphors), 2x)


FFVI (3x)


Silent Hill (Original, 3x)


SoM2 (Original, 2x)

PHEW!

byuu, I'll try and get the info on how these work sent to you soon!

One more thing I'd quite like to add is an optional highlight bloom, a bit like this (note, this is only a mockup):



Well, let me know what you all think!

Thanks!

This. I want this to be as widely available as blargg's NTSC filter is now.

it all looks great!

also the bloom highlight looks good.

Looks very nice.

Would there be a difference in speed between big and small phosphors?

That's a huge improvement.

If you could add the highlight and lowlight bloom effect i think it will be perfect.

Glad you guys like it!



They're more or less the same. I think the small phosphors may be slightly slower, ~5% or so, but not a huge difference. I'll check it properly tonight and get back to you.

EDIT: Ok, so currently the smaller phosphor version of the 3x algortihm takes ~12% longer than the big phosphor version. I'm not much of a programmer though, so that doesn't necessarily mean it would be the case for a real implementation of the algorithms.

Ok guys, I've made a video of the 2x algorithm in action using Secret of Mana (via bsnes of course!) and uploaded it here. Set it to fullscreen and enjoy!
NOTE: It's ~50MB, but I've used trafficshare so it's a direct download, meaning you don't have to wait like you normally do with rapidshare.

I was just curious to see how it would look in motion. The water effects look especially pleasing