OK...
Nooo! This is a complete confusion. Edge enhancement=detail degradation, because more prominent detail obliterates lesser detail. After many months of dispute you will not accept this, so we'll have to agree to disagree. Except you are wrong. Trying to reach common ground, if the picture lacks detail a bit of user added edge enhancement can make it more engaging or appealing, but it's no substitute for the real thing. Agreed?
I'm afraid I'm going to disgree with you here once again as all my experience with HD video and Stills says otherwise.
In the link below is an industry standard resolution chart.
The I've copied 1/2 of it and repasted it onto the right hand side to which I've applied 30% unsharp mask.
Maybe you'd like to show me where detail has been lost using this chart?
In fact I would argue otherwise.
I've added an arrow onto the bottom of the chart pointing to one of the grey scale charts. In my eyes I can see MORE fine grain detail on the right hand side (the sharpened side) than I can on the left (the unsharpened side).
As I haven't added any grain, it can only be the greater edge detail in the sharpened picture enhancing the fine grain detail thats already there in the picture (or more correctly making it stand out more through the shraper edge detail).
Picture is here:
Once navigated, select the magnifying glass icon to make it full screen.
I'm dissapointed that with new HD cameras and the ability to transmit quality HD via Freesat and Freeview that we are are not getting what was promised a year or so ago.
I'm now finding that some programmes that were filmed with HD cameras and transmitted on SD channels are as good as HD recorded programmes transmitted on HD channels, well theres not much in it.
I'm afraid I'm going to disgree with you here once again as all my experience with HD video and Stills says otherwise.
In the link below is an industry standard resolution chart.
The I've copied 1/2 of it and repasted it onto the right hand side to which I've applied 30% unsharp mask.
Maybe you'd like to show me where detail has been lost using this chart?
In all the places where fine detail is left to be almost inferred by the eye because it's so subtle. A mouseover version of your chart brings this out. (Moving the mouse above & below the bottom of the image is the most convenient method of changing it--too tired to sort out the horizontal now.)
What I'm looking at here are the converging lines. The added sharpness makes them more distinguishable from each other without doubt, but each one becomes more crooked, and the delicate relationships between them precisely as converging become harder to see. A couple of examples.
I've added an arrow onto the bottom of the chart pointing to one of the grey scale charts. In my eyes I can see MORE fine grain detail on the right hand side (the sharpened side) than I can on the left (the unsharpened side).
As I haven't added any grain, it can only be the greater edge detail in the sharpened picture enhancing the fine grain detail thats already there in the picture (or more correctly making it stand out more through the shraper edge detail).
Yes indeed, but it was not without cost elsewhere as I hope I've shown. Which brings me back to what I proposed as common ground--"if the picture lacks detail a bit of user added edge enhancement can make it more engaging or appealing, but it's no substitute for the real thing." There are compromises, sometimes preferable compromises, involved in compensating for lack of information. Can we agree on this now?
I agree with you, White-Knight. The key point for me is the phrase "because it stands out more". The sharper the picture the more detailed it appears. This is a human perception issue.
This does not actually mean that there is more detail just that the brain thinks there is more detail. I am not just playing with words here. You can get a human to look at the same scene shot with two different cameras & they may perceive one as sharper & more detailed than the other. However when you actually shoot a test card then the other camera can be proved to resolve more detail. A good example is between the Canon 5D Mk II & a Sony EX1. The picture of the former looks sharper whereas the latter is actually more detailed.
We're splitting hairs over the meaning of words here, but the "sharpened" image from White-Knight clearly has the higher frequencies boosted, and nothing has been lost as far as I can see. The ringing isn't too bad, though I don't think I'd choose to watch TV with this enabled!
The problems seen on broadcasts have little to do with a simple unsharp mask performed on an uncompressed test chart. Both the denoising and the MPEG encoding cause a kind of threshold effect: high frequencies are removed if they are low amplitude, and retained if they are high amplitude. Combined with sharpening, this has the effect of boosting high amplitude high frequencies, and removing low amplitude high frequencies. This gives an artificially-sharp-but-plastic look (to my eyes anyway).
Sharpening can also introduce ringing/halos on already sharp edges.
That said, I think the right amount of careful sharpening is fine and desirable - it helps some MPEG encoders to preserve relevant details - i.e. those that would be missed most. Plus it actually looks nice - a raw completely unsharpened image looks very soft indeed - I doubt most people have ever seen one. Every camera uses some aperture correction: http://www.adamwilt.com/TechDiffs/Sharpness.html
This does not actually mean that there is more detail just that the brain thinks there is more detail.
I agree. I never said it could introduce more detail just that greater edge detail increases the contrast between detail and surrounding areas making it more visible to the eye. Thus a sharper picture appears more detailed even though it actually has the same detail as an unsharpened picture, you can just see it more easily.
We're splitting hairs over the meaning of words here, but the "sharpened" image from White-Knight clearly has the higher frequencies boosted, and nothing has been lost as far as I can see. The ringing isn't too bad, though I don't think I'd choose to watch TV with this enabled!
The problems seen on broadcasts have little to do with a simple unsharp mask performed on an uncompressed test chart. Both the denoising and the MPEG encoding cause a kind of threshold effect: high frequencies are removed if they are low amplitude, and retained if they are high amplitude. Combined with sharpening, this has the effect of boosting high amplitude high frequencies, and removing low amplitude high frequencies. This gives an artificially-sharp-but-plastic look (to my eyes anyway).
Sharpening can also introduce ringing/halos on already sharp edges.
That said, I think the right amount of careful sharpening is fine and desirable - it helps some MPEG encoders to preserve relevant details - i.e. those that would be missed most. Plus it actually looks nice - a raw completely unsharpened image looks very soft indeed - I doubt most people have ever seen one. Every camera uses some aperture correction: http://www.adamwilt.com/TechDiffs/Sharpness.html
Cheers,
David.
In fairness I have to point out David that I downloaded the test chart from the internet and it had obviously been photographed (poorly) as it had Moire (ring like patterns) already on it which is why the image isn't as clear as it could be.
There is no real additional interference added by the sharpening, maybe some of it stands out a little better as a result, but the problems with ringing seen in the chart stem from the fact that the chart was far from perfect to begin with.
Sharpening especially at modest levels doesn't result in intereference. In fact even harsh sharpening doesn't, it causes aretefacing which is entirely different.
The reference to lines being crooked was an attempt to give an example of what I meant by "places where fine detail is left to be almost inferred by the eye because it's so subtle". Though the following way of putting it has its problems I know, it's sort of the opposite of adding anti-aliasing: certain parts of the image stand out more now, but the difference between them and other pixels is increased such that you are less likely to notice the dimmer pixels, and the overall shape of the line is less clear to the eye, even though no pixels have changed position. Sorry if that sounds a bit vague, but here's an example with a huge amount of unsharp mask added that I hope makes clear what I mean.
There is certainly distortion here. The lines looks more like like gnarled twigs rather than straight lines. Now obviously no-one would apply this amount of sharpness in real life (I hope!), but I can see a much subtler version of it going on in your 30% unsharp mask. That's what I'm talking about, anyway.
Two final points--I realised that this exercise is somewhat different from actual sharpness in real video images, as 2Bdecided points out, but it seemed worth pursuing anyway. And yes, as I said, a bit of user added sharpness can make a picture more acceptable but it's not a substitute for the real thing. 2Bdecided's point that every camera uses it to some extent shows one (I ) shouldn't be too absolutist about it though.
Two final points--I realised that this exercise is somewhat different from actual sharpness in real video images, as 2Bdecided points out, but it seemed worth pursuing anyway. And yes, as I said, a bit of user added sharpness can make a picture more acceptable but it's not a substitute for the real thing. 2Bdecided's point that every camera uses it to some extent shows one (I ) shouldn't be too absolutist about it though.
I agree. Most digital cameras shoot soft by default though, its well known.
It seems the BBC have a habit of outputting soft pictures.
A good example by way of contrast was tonight's River Monsters on ITV HD.
Picture was razor sharp in places and all the better for it.
Detail was sublime at times and good at others.
A few of the most memorable shots were of the sea plane including one where it was in dock at the lakes edge and where it was so clear you actually felt drawn into the picture almost like reality.
I do agree you can overdo sharpness, I don't believe I ever said otherwise. However, for the vast majority of material razor sharp produces the better HD both for clarity visibility of detail. Of course there are exceptions where for mood reasons it may be permissible to deviate from sharpness to soft or even grainy.
I would also re-iterate that the test pattern I put up above was a download and already poorly reproduced with visible moire (which is why its on both the sharpened and unsharpened sides), although I do take your point in your example albeit the amount of sharpening you've applied is extreme. I would suggest what you're seeing is not bending of the lines but stepping which is there anyway but just far more visible in the example with the extreme sharpening applied.
Just out of interest, which of the above 200% pictures did you prefer?
I would also re-iterate that the test pattern I put up above was a download and already poorly reproduced with visible moire (which is why its on both the sharpened and unsharpened sides), although I do take your point in your example albeit the amount of sharpening you've applied is extreme. I would suggest what you're seeing is not bending of the lines but stepping which is there anyway but just far more visible in the example with the extreme sharpening applied.
Fair enough--I don't mind whether it's called bending, stepping, jaggies or whatever, it's clearly a distortion of the image--to the point where the straight lines look anything but, and the middle lines in the the cut-out in #210 have mostly become one very thick one.
Just out of interest, which of the above 200% pictures did you prefer?
Well, of course I prefer the unsharpened one , i.e. the one on the right. And for the same reason really: the straight lines look a bit less straight on the sharpened one. I know the white triangles and other white against grey detail stand out a bit better, and this is a still graphic not a real video image, etc., but all the same, that's my preference, even though your example uses a more realistic/reasonable amount of sharpening than my extreme example.
Comments
I'm afraid I'm going to disgree with you here once again as all my experience with HD video and Stills says otherwise.
In the link below is an industry standard resolution chart.
The I've copied 1/2 of it and repasted it onto the right hand side to which I've applied 30% unsharp mask.
Maybe you'd like to show me where detail has been lost using this chart?
In fact I would argue otherwise.
I've added an arrow onto the bottom of the chart pointing to one of the grey scale charts. In my eyes I can see MORE fine grain detail on the right hand side (the sharpened side) than I can on the left (the unsharpened side).
As I haven't added any grain, it can only be the greater edge detail in the sharpened picture enhancing the fine grain detail thats already there in the picture (or more correctly making it stand out more through the shraper edge detail).
Picture is here:
Once navigated, select the magnifying glass icon to make it full screen.
I'm now finding that some programmes that were filmed with HD cameras and transmitted on SD channels are as good as HD recorded programmes transmitted on HD channels, well theres not much in it.
In all the places where fine detail is left to be almost inferred by the eye because it's so subtle. A mouseover version of your chart brings this out. (Moving the mouse above & below the bottom of the image is the most convenient method of changing it--too tired to sort out the horizontal now.)
http://dl.dropbox.com/u/5159779/test_pattern_whole.htm
What I'm looking at here are the converging lines. The added sharpness makes them more distinguishable from each other without doubt, but each one becomes more crooked, and the delicate relationships between them precisely as converging become harder to see. A couple of examples.
http://dl.dropbox.com/u/5159779/test_pattern_horizontal_detail.htm
http://dl.dropbox.com/u/5159779/test_pattern_circle_detail.htm
In the latter case they look more staircased as they approach the centre of the pattern, which is obviously not the required result.
Yes indeed, but it was not without cost elsewhere as I hope I've shown. Which brings me back to what I proposed as common ground--"if the picture lacks detail a bit of user added edge enhancement can make it more engaging or appealing, but it's no substitute for the real thing." There are compromises, sometimes preferable compromises, involved in compensating for lack of information. Can we agree on this now?
The problems seen on broadcasts have little to do with a simple unsharp mask performed on an uncompressed test chart. Both the denoising and the MPEG encoding cause a kind of threshold effect: high frequencies are removed if they are low amplitude, and retained if they are high amplitude. Combined with sharpening, this has the effect of boosting high amplitude high frequencies, and removing low amplitude high frequencies. This gives an artificially-sharp-but-plastic look (to my eyes anyway).
Sharpening can also introduce ringing/halos on already sharp edges.
That said, I think the right amount of careful sharpening is fine and desirable - it helps some MPEG encoders to preserve relevant details - i.e. those that would be missed most. Plus it actually looks nice - a raw completely unsharpened image looks very soft indeed - I doubt most people have ever seen one. Every camera uses some aperture correction: http://www.adamwilt.com/TechDiffs/Sharpness.html
Cheers,
David.
I agree. I never said it could introduce more detail just that greater edge detail increases the contrast between detail and surrounding areas making it more visible to the eye. Thus a sharper picture appears more detailed even though it actually has the same detail as an unsharpened picture, you can just see it more easily.
In fairness I have to point out David that I downloaded the test chart from the internet and it had obviously been photographed (poorly) as it had Moire (ring like patterns) already on it which is why the image isn't as clear as it could be.
There is no real additional interference added by the sharpening, maybe some of it stands out a little better as a result, but the problems with ringing seen in the chart stem from the fact that the chart was far from perfect to begin with.
Sharpening especially at modest levels doesn't result in intereference. In fact even harsh sharpening doesn't, it causes aretefacing which is entirely different.
How can sharpness bend lines?
I've picked out one instance of the detail you focused on above and have blown it to 200% here:
http://img547.imageshack.us/img547/3839/test2i.jpg
Maybe you'd like to point out where lines have been bent?
To bend line pixels would have to be shifted. All sharpening does is give a harder edge to the pixels it doesn't shift their position.
The reference to lines being crooked was an attempt to give an example of what I meant by "places where fine detail is left to be almost inferred by the eye because it's so subtle". Though the following way of putting it has its problems I know, it's sort of the opposite of adding anti-aliasing: certain parts of the image stand out more now, but the difference between them and other pixels is increased such that you are less likely to notice the dimmer pixels, and the overall shape of the line is less clear to the eye, even though no pixels have changed position. Sorry if that sounds a bit vague, but here's an example with a huge amount of unsharp mask added that I hope makes clear what I mean.
http://dl.dropbox.com/u/5159779/test_pattern_horizontal_detail_xtra_sharp.htm
There is certainly distortion here. The lines looks more like like gnarled twigs rather than straight lines. Now obviously no-one would apply this amount of sharpness in real life (I hope!), but I can see a much subtler version of it going on in your 30% unsharp mask. That's what I'm talking about, anyway.
Two final points--I realised that this exercise is somewhat different from actual sharpness in real video images, as 2Bdecided points out, but it seemed worth pursuing anyway. And yes, as I said, a bit of user added sharpness can make a picture more acceptable but it's not a substitute for the real thing. 2Bdecided's point that every camera uses it to some extent shows one (I
I agree. Most digital cameras shoot soft by default though, its well known.
It seems the BBC have a habit of outputting soft pictures.
A good example by way of contrast was tonight's River Monsters on ITV HD.
Picture was razor sharp in places and all the better for it.
Detail was sublime at times and good at others.
A few of the most memorable shots were of the sea plane including one where it was in dock at the lakes edge and where it was so clear you actually felt drawn into the picture almost like reality.
I do agree you can overdo sharpness, I don't believe I ever said otherwise. However, for the vast majority of material razor sharp produces the better HD both for clarity visibility of detail. Of course there are exceptions where for mood reasons it may be permissible to deviate from sharpness to soft or even grainy.
I would also re-iterate that the test pattern I put up above was a download and already poorly reproduced with visible moire (which is why its on both the sharpened and unsharpened sides), although I do take your point in your example albeit the amount of sharpening you've applied is extreme. I would suggest what you're seeing is not bending of the lines but stepping which is there anyway but just far more visible in the example with the extreme sharpening applied.
Just out of interest, which of the above 200% pictures did you prefer?
Fair enough--I don't mind whether it's called bending, stepping, jaggies or whatever, it's clearly a distortion of the image--to the point where the straight lines look anything but, and the middle lines in the the cut-out in #210 have mostly become one very thick one.
Well, of course I prefer the unsharpened one
PS must look out for River Monsters!