Megapixel race still on

Discussion of all digital SLR cameras under the Minolta and Konica Minolta brands
Forum rules
No more than three images or three external links allowed in any post or reply. Please trim quotations and do not include images in quotes unless essential.
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Megapixel race still on

Unread post by agorabasta »

Greg Beetham wrote:... the ragged transitions that occur when downsizing the jpeg for the web, ...
No, Greg, that problem is fundamental.

First, all the gamma-emphasized formats with the 8bit colour must produce some posterisation on smooth colour transitions, especially in the brighter parts of the image where that gamma pre-emphasis makes digital levels too scarce.
Second comes the encoding of JPEG format, that is not RGB, it's Y-Cr-Cb; and that Y (luma) is only 8bit even before compression. Then the chroma Cr Cb are normally (4:2:2) encoded with half spatial bandwidth. And also the nature DCT JPEG compression is to discard small changes in the original value until the accumulated change becomes too large.

Now if you downsize for web, you not only overcompress causing banding in DCT, you also average out the noise that was originally providing some dithering that prevented posterisation.

So the only way to keep smooth colour transitions in overcompressed downsized JPEGs is to use dithering by adding noise to the original.
User avatar
Greg Beetham
Tower of Babel
Posts: 6117
Joined: Sun May 27, 2007 3:25 pm
Location: Townsville, Qld. Australia
Contact:

Re: Megapixel race still on

Unread post by Greg Beetham »

Ok agorabasta that sounded plausible I don’t know that much about how JPEG’s are constructed, or what the DCT is, but what are the chances of avoiding ragged transitions in a heavily gamma looking sky photo no matter what you do when trying to get to web size? I almost found a way but like I say I had to keep at around ½ Mb in size, even then there was a very slight rough (fragmented) transition but you had to look at an angle and know where to look…but it bordered on being acceptable. Too me the sheer amount of reduction required for a large MP camera to get down to web size makes me think that those files would be even more difficult to deal with than a smaller sensor camera, in that type of situation.
I would like to be shown that that idea is wrong of course, but I’ve got my doubts.
Thinking about it you would think that the seeing as how the linear size is being diminished at the same time as the file content is being compressed a certain amount of dithering would take place anyway; maybe one reason is you can’t vector a smooth transition, only things with edges work properly?
What about .bmp conversion pre reduction? I don’t know if it’s possible to convert back again though, without creating an even worse mess, I think .bmp is constructed off the screen and not off the image file so that means all the exif is gone if that’s the case.
Greg
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Megapixel race still on

Unread post by agorabasta »

Greg,

BMP is kind of 'screen pixels' image, but win/mac screens are also gamma-corrected. So the BMP's are also gamma-corrected, just as JPEG is. The mac gamma is slightly less steep, it's 1.8; while the win gamma is 2.2, just as the JPEG is. So - no go. Just wait for 30bit colour (10b per channel RGB).

I've been following that old thread where everybody complained of poor sky transitions, so I then went and took some test pics with a polariser (Sigma CPL for the DP series, 19f2.8 Sigma on Nex7).
Below are two versions of the same sample, one downscaled to 50% in Lr4, the other is not; then they both were exported at 1000p width 250kB size limit. To me, there's absolutely no diff between them; furthermore, if you see any sky banding, it's your monitor problem (mine is 10b per channel internally and not overcalibrated into banding). But there still is another possibility that banding may be introduced by the web browser/viewer being overzealous at colour management.
Attachments
DSC06381-1.jpg
(247.2 KiB) Downloaded 3489 times
DSC06381-2.jpg
(243.86 KiB) Downloaded 3489 times
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Megapixel race still on

Unread post by agorabasta »

Actually there is some diff between the two samples, but it's quite random over the smooth areas and is stronger where images differ most in the spatial spectrum. Here it is scaled into the viewable range (original diff is up to 12 of the 0-255 scale), so downscaling itself cannot introduce banding -
Attachments
DSC06381-diff.jpg
(219.86 KiB) Downloaded 3487 times
User avatar
Greg Beetham
Tower of Babel
Posts: 6117
Joined: Sun May 27, 2007 3:25 pm
Location: Townsville, Qld. Australia
Contact:

Re: Megapixel race still on

Unread post by Greg Beetham »

Agorabasta I think the first one looks a little more contrasty than the second, the sky isn’t bad in both but I can see a faint representation of the ragged sky problem and not only that I think there is a couple of places where they are very similar (the second photo is harder to see it in for comparison purposes). The last is too pixelated for me, I’m not sure what I’m supposed to see with that.
I have two computers and the sky in both of those photos look better on the Viewsonic screen than the Samsung, it’s still slightly ragged looking but quite a bit harder to see.
I’d have to say though overall you probably wouldn’t notice it at all on either screen if you weren’t looking for it.
Greg
Ps where is everyone? has approaching winter scared them off?
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Megapixel race still on

Unread post by agorabasta »

Greg Beetham wrote:The last is too pixelated for me, I’m not sure what I’m supposed to see with that.
It's the difference between the two above, multiplied by about x20 factor.

If you see any irregularities in the sky on those two samples, it comes exclusively from your machines/monitors. Chances are, your Samsung is a 6bit TN screen. Viewsonic make some half-decent IPS (LG-made HDTV 16:9 screens), but no 10bit screen, only 8bit (means posterisation with colour management). All Samsung-made 10bit IPS screens seem to go into Dell monitors.

You make check your monitors for banding using tests from this page http://tft.vanity.dk/
User avatar
Greg Beetham
Tower of Babel
Posts: 6117
Joined: Sun May 27, 2007 3:25 pm
Location: Townsville, Qld. Australia
Contact:

Re: Megapixel race still on

Unread post by Greg Beetham »

Thanks agorabasta, I guess you gets what you pays for, or in my case what you can afford, (plus what they have in stock when you want one) what is a decent monitor in your view?
Greg
agorabasta
Viceroy
Posts: 1198
Joined: Wed Oct 22, 2008 7:41 pm

Re: Megapixel race still on

Unread post by agorabasta »

Greg,

I'm still using a 3-year old Dell U2410, since then they made the U2411-U2412. All have IPS panels of 1920x1200 16:10 format, but only U2410 is 10bit and can be 'calibrated' by hand using internal controls. Some earlier U2410 had some problems if both hue and saturation controls in the monitor were adjusted together. No problems since 2-year ago manufacturing date. The U2410 is still the best of affordables, it would seem.

The cheapest IPS of workable size are made over the e-IPS panels by LG. They are 23" 1920x1080 16:9 panels, 8-bit only. Many brands use the same LG panel, they are Asus, Viewsonic, Fujitsu, Philips etc.

The prices differ about twofold between those Dell U2412 and the cheaper 23" makes, while U2410 is about 1.5 times more than U2412. If you need to shop online, just search for 'Dell IPS' or '23" IPS' or whatever size you prefer with 'IPS' term. Btw, 27" screens are much more comfortable to work with, if you can find an older 1920x1200 model, and not anything of higher res. A good old model was a Dell 2709W, not an IPS, but a very decent VA screen.

Here's a good site with professionally made monitor reviews - http://www.prad.de/en/index.html
Post Reply

Who is online

Users browsing this forum: No registered users and 7 guests