byuu's message board

Yup, that's what I said. A few. It doesn't matter though cause the console doesn't make use them.

Admittedly a PAL CRT does have less noticeable dark spaces between the scanlines since they are closer together, but that doesn't help with emulators since the strength of dark scanlines is user configurable.

So again, why play them in emulators? Surely having the game run much slower than the game designers intended far outweighs any benefits, if they even existed? Not to mention a skewed aspect ratio caused by the extra black scanlines shoehorned into the signal to make it a valid PAL one.

I'm not saying it's something I won't look at, it's just near the bottom of the list as I can't quite fathom why anyone would want to play a PAL game, other than for pure nostalgia reasons (remembering the game running slower and looking a certain way).

i have a pal TV, so i've never seen the correct NTSC aspect ratio ever.
Basically anything in NTSC display mode looks wrong to me.(eventhough it should be accurate)

Isn't nostalgia the reason for these kinds of filters anyway?

Progress:

Proper scanlines (NTSC and PAL).

Less aliasing.

Better brightness mapping.

Subtle gamma difference from unfiltered image.

Issues:

Inaccurate.

Still some aliasing left.

Examples:

CRT + Blargg's NTSC Filter:

PAL aspect ratio and scanlines:

Trinitron style phoshpor mask:

Once I'm comfortable with the method I will proceed to optimize it and process a video.

Edit: For the sake of comparison, here are some photographs:
http://scanlines.hazard-city.de/sf3_dc_rgb_lr_131.jpg
http://scanlines.hazard-city.de/sf3_dc_rgb_lr_121.jpg

Notice that the pictures are slightly underexposed and that allows for more phosphor definition.

Here's something you can compare them to:

I don't think your latest images represent a CRT faithfully Dolls, but you've said yourself that it is inaccurate, so...

Btw, which language are you writing your filter in? Xythen is using Matlab I believe.

It is more accurate than the previous revisions, though.
I'm prototyping in Java.

When we get the phosphor3x? I going withdraws now.

His images are extremely accurate,
The problem is he isn't simulating the phycovisuall effects, which is why the checkerboard is so distracting.

Thank you but, to be fair, the simulation is still ways off... and I don't think it is feasible to have a proper one running in realtime along with bsnes, if it's not in pixel shader form at least.

I finally got ahold of a SNES to test on my old TV set and got to this a few days ago:

I realized the results can look dramatically different on an LCD display than on my CRT monitor, and I haven't had the chance to perform thorough testing to eliminate the extreme aliasing that LCD displays show, I'll try and switch to another sampling strategy when I have the time.

All the blending operations were computed taking into account that your display runs in a more or less faithful sRGB colorspace, please do test in such mode if available, since the gamma can vary wildly otherwise.

Wow, it's really starting to look great now DOLLS

I especially like the colour blooming around the edges of bright objects, like the white writing in the zelda pic. My original 3x version achieved something similar. The colour streaking is really cool too, though perhaps a little overdone. It seems to make the Mario image a bit more blurred than it should be.

I really like the subpixel structure you've managed to achieve, looks really convincing! I'd imagine it's really expensive to compute though. But like you said, there might be hope for some sort of GPU implementation.

The main thing it needs now is proper scanline simulation. A really good test for this, as was pointed out by tetsuo55 a few pages back, is Silent Hill for the PSX since it relies on dithering between the scanlines to produce the fog effects. Here's a good pic for you to work with (note that although the PSX renders it to 320x224 it should still display as 4:3, so some aspect correction is also needed!)




Just as a side note, isn't the main point of these CRT simulation efforts to reproduce a CRT-style effect on newer display technologies (LCDs etc.)? If you're using a CRT monitor, doesn't it kind of defeat the point? Stick the NTSC filter on with some scanlines and you're basically there, surely?

Thanks!


This is positively the result of gamma-corrected manipulation, It would seem to me that it tends to favor light tones instead of dark ones (the result of no gamma correction), but it's all conforming to the W3C spec.


Yes, but that's quite close to the way my shitty old TV displays things at the moment . I'm planning on making all these parameters tunable, so that shouldn't be a worry.


It is expensive as you said, but right now it has to be calculated only once per resolution change. I'm working on reducing aliasing even further, there's just so much high frequency content in there that it has become a bit costly in development and running time.


Thank you. I've been thinking on a method of manipulating scanline thickness that would be cheap enough, but haven't got too far.


Hehe, I think this is an oversimplification, any computer screen is leaps and bounds ahead in terms of resolution and precision from a TV set, specially one from the 90's. But yes, the point stands, unfortunately I don't have continuous access to a LCD screen at the moment.
I'm planning on adding subpixel blending/smoothing in the spirit of ClearType in the future, to see if it helps at all.

Even though my technique is now really close to my own TV set, I find the result of your filters more pleasing to the eye (and close to what a SNES looks like on a really big trinitron screen), so I don't consider the development of either mutually exclusive, as we're covering slightly different grounds. But I wonder if you would be seeking convergence and share insights, as tetsuo55 suggested some posts ago.

I'm not entirely sure I'm displaying the correct amount of scanlines, I based my calculations on the assumption that a NTSC set is expecting to display 480 lines, so if the signal received is composed of 224, it would end up displaying 256 'empty' scanlines and sharing the content of the signal among the rest. Is this correct?.

This might be a property of the NTSC signal, instead of the TV screen.

Not quite. The way I understand it, the TV has to receive a valid NTSC signal so it is the SNES which has to convert the 256x224 image into the correct format and transmit it to the TV. The TV paints even then odd lines to the screen (interlacing) which the SNES exploits to create a progressive image, just setting all odd lines to black. The extra lines to make up 480 are padded with black, but I'm not sure if the TV recognises this and puts it in the overscan region, or if there are actually black bars on screen...

testsuo explained it with some example pictures here, as well as a few pages back:



Again, I'm not sure how the TV deals with overscan regions, which I guess is why there tends to be disagreements over what the output aspect ratio should be.

On an old-school CRT, all that happens is that the electron beam sits on horizontal and vertical rails. It moves along both these rails at a certain speed determined by the signal it receives, so that by the time it has finished scanning a horizontal line and returned to the left, it will have moved down enough to start drawing the next line (2 lines in interlaced mode, 1 line in progressive). In interlaced mode it first scans all the even lines, then when it reaches the end of this field (one half of an interlaced frame) it intentionally scans only half of the last line, so that when it moves back up, it'll start half a line too early and end up filling all the odd lines.

Interlacing is good enough for human eyes because the afterglow of the even and odd lines is enough that they end up blending together. One thing I don't know is whether you also see the odd lines of the second field of a frame blend together with the even lines of the next field. I'd imagine so, but maybe the wait between consecutive frames is longer than the one between fields?

The images on my blog contain ALL the visable video data drawn on a perfectly calibrated CRT TV.

This TV would probably have masks (all CRT tv's too) that would cut off most of the black parts(the black parts are in the overscan area, behind the mask) This also means that for the parts with color, the aspect ratio is theoretically perfect.