Rutland Church

[Stevewebb]

Re-visited an image taken last year to give it a bit more of a mono treatment.

Any thoughts?

[Barry]

I like this, lots of Doom and Gloom here ;D. The pathway leads you up to the church well and the sky is very atmospheric. The church stands out well against the background. But I did wonder if you have gone a bit too far with the contrast or sharpening especially around the rails on church roof.

[jeeperman]

I like it as well Steve and also agree with Barry's observation.

[The Wirefox]

Very Gothic Steve. I like this it is well composed but is exhibiting some of the issues I have discussed in my GIMP tutorial for the Cannock Wolf conversion method. There is a a fair amount of noise in the sky and as Barry points out there are problems with edge artifacts on both on the church and in the clouds. Chris has been having similar problems with his conversions. I am beginning to suspect this is an issue that can be attributed to heavy application of contrast at anyone time and it seems to be compounded by the fact that the contrast adjustment in the wolf method and SEP is very similar in method to the means of adjusting exposure using a layer method. It may be better to do this type of conversion in stages a bit at a time using non-lossy file formats. For instance either applying the wolf method in 3 increments rather than in one go and same for SEP. Introducing a Guassian blur layer or a monochromatic grain layer may also alleviate the issues with the sky. Daves conversions do not seem to suffer from this though and I can't figure out why from his tutorial.

I am going to have a go at this myself to see if I can isolate the problem

[Stevewebb]

Thanks for the comments. All very valid of course.

I think my error on this one was applying a high pass sharpening as it seems to be that part that added a lot more noise into the sky.

I also seem to be struggling when re-sizing images to display on the web. My normal routine is to crop the full size image down to 6 x 4.5 in then apply a low amount of sharpening (100%, 1 pixel, threshold 10), then resize to 1000 px wide and apply the same sharpening again. I think that this doesn't mix well when have already applied a high pass to the full size image.

The re-sizing also seems to make the contrast seem a lot more severe as the top of the church rails do not look that bad in the full size image.

I did suffer from edge artifacts around the church when processing this but it manifested itself more in the form of CA so I put it down to using the kit lens in challenging light conditions.

For interest here if my workflow for this image in CS5-

Open original image in raw and process for the church, then open as a smart object in CS5
Duplicate the layer as a smart object via copy,
Go back to the first layer and double click to open in raw again and process for the sky.
So I now have a light layer on top and a dark layer underneath.
On the light layer make a selection around the church and output the selection to a layer mask so it brings the dark sky through.
With the gradient tool put a ND gradient left to right to darken the left side of the church.
With the gradient tool put a ND gradient bottom to middle to darken the path at the bottom of the image.
Add a new layer in overlay blend mode and fill with 50% grey.
Use the brush tool set at 15% opacity to paint in light and dark areas gradually (white=light, Black=dark)

[Barry]

Hi Steve, whenever I edit a image from RAW or Jpeg, I always save it at its native size as a Tiff file. Then to upload here, I just open the edited Tiff file image and just resize to 1000 px along top for landscape or 800px high for portrate, and save using quality 9 or 10, and upload, I never sharpen or add any more contrast.

[Stevewebb]

Interesting, thanks Barry. I will have a play and look at the difference. That is if smugmug will display TIFF files?

[Barry]

Jan 23, 2012 17:37:54 GMT Stevewebb said:

Interesting, thanks Barry. I will have a play and look at the difference. That is if smugmug will display TIFF files?

Sorry I forgot that bit, after I have resized them, they are saved as Jpeg quality 9 or 10 before uploading to Flikr. I don't upload Tiff files.

[Stevewebb]

Ah Hah ;D

[The Wirefox]

Steve why do you resize for SmugMug? I just upload full size .jpg (you can not upload tiffs for viewing in Smug). I then go to share / image link tab then copy XL size (1024px width) and paste it between the [IMG/] tags on the forum. You can prevent viewers from seeing your original size images in the SmugMug gallery preferences.

There seems to be nothing in your PP workflow that would cause an increase in noise (sharpening aside). I think the only solution is probably to bracket. Expose one shot for the sky and one shot for the subject and then open as separate layers painting through to reveal the correctly exposed sky. Also expose to the right. Bunch the histogram up to the right axis so that you are capturing the maximum amount of pixel data. Set blinkies to check you are not clipping. That way you capture maximum of data for the sky and maximum data for the subject. The less data captured the higher the noise levels (not as simple as that really but unless you want to get your head around this...

Most of the noise in an image is just photon noise (shot noise) from the light itself. In a Poisson distribution, the variance (the square of the noise) is equal to the mean. So with each additional stop of light captured, the square of the noise doubles, which means the noise itself (the std dev) only goes up by a factor of sqrt (2). This is why SNR improves with more exposure (2x the light means only 1.4x the noise), which is why ETTR works. A simple but quite accurate noise model for digital capture is just a square root function:

noise (x) = sqrt (A*x + B)

where

x is the average signal (say, in a normalized range from [0,1])

A is determined by the size of the pixel and the chosen ISO (A*x is the photon noise), and

B is the noise floor of the sensor, which is independent of the # of photons you captured

With a perfect noise-free sensor, B is zero, but you still have A -- darn it!

With no light (taking pictures of the black cat in a cave, or in my case, taking pictures with a lens cap on with the lights off), you have no photon noise since x is zero, but you still have B because sensors aren't perfect -- darn it!

As you can see, as x grows, so does noise (x). You can easily measure noise by taking a picture of a uniform area and measuring the mean and variance of the pixels. Do this for several exposure levels to determine a set of data points. They should lie on a straight line. The slope of that line is A, and the y-intercept of that line is B. Voila, the fundamental noise properties of your sensor! - From 'Optimising Exposure' - The Luminous Landscape


So, to illustrate the importance of ETTR in a digestible form;

- The brightest stop = 2048 tonal values
- The next brightest stop = 1024 tonal values
- The next brightest stop = 512 tonal values
- The next brightest stop = 256 tonal values
- The next brightest stop = 128 tonal values
- The next brightest stop = 64 tonal values
- The darkest stop = 32 tonal values

(assumes a modern DSLR capturing 10 stop dynamic range)

There is of course a complication. With the methods being used in recent b/w landscape images appearing on FSC we are going for high contrasts in the clouds. This means that we have very dark areas adjacent to very light areas and increasing that tonal difference artificially. if we expose for the light areas of the sky the darker areas will be underexposed and contain noise (leading an increase of noise and posterisation when sharpened) so we may need to bracket exposure for the darker clouds exposing to the right(ETTR), the subject (ETTR) and the light areas of the sky (ETTR). thus armed we can select the best parts of the exposure in each layer. In this way, because we have captured as much data as possible by ETTR we can darken and lighten certain areas for effect and minimise the amplification of noise. Sharpening will amplify noise so it is best to use a method that allows selective sharpening or edge sharpening.

So if you are in a hurry or conditions dictate that only a single shot is possible you are going to have to live with compromise. This will not be such a massive problem SOOC with minimal PP but with the heavy processing that is trundling away in SEP and multi-layer methods (or even heavy leaning on levels or curves) you will introduce noise and selective Guassian blurring and selective sharpening may the your only partial get out. For problems with banding or posterisation the introduction of a monochrome grain layer may also help.

Note: You can see why Jeroens long exposure shots get around this problem because motion blur occurs in the sky area eradicating the noise and potential posterisation.

[Barry]

Jan 23, 2012 20:30:23 GMT The Wirefox said:

So, to illustrate the importance of ETTR in a digestible form;

- The brightest stop = 2048 tonal values
- The next brightest stop = 1024 tonal values
- The next brightest stop = 512 tonal values
- The next brightest stop = 256 tonal values
- The next brightest stop = 128 tonal values
- The next brightest stop = 64 tonal values
- The darkest stop = 32 tonal values

(assumes a modern DSLR capturing 10 stop dynamic range)

This is something that I was reading about over Christmas, as I did not really understand it before.

[cannockwolf]

Jan 22, 2012 21:52:14 GMT The Wirefox said:

Very Gothic Steve. I like this it is well composed but is exhibiting some of the issues I have discussed in my GIMP tutorial for the Cannock Wolf conversion method. There is a a fair amount of noise in the sky and as Barry points out there are problems with edge artifacts on both on the church and in the clouds. Chris has been having similar problems with his conversions. I am beginning to suspect this is an issue that can be attributed to heavy application of contrast at anyone time and it seems to be compounded by the fact that the contrast adjustment in the wolf method and SEP is very similar in method to the means of adjusting exposure using a layer method. It may be better to do this type of conversion in stages a bit at a time using non-lossy file formats. For instance either applying the wolf method in 3 increments rather than in one go and same for SEP. Introducing a Guassian blur layer or a monochromatic grain layer may also alleviate the issues with the sky. Daves conversions do not seem to suffer from this though and I can't figure out why from his tutorial.

I am going to have a go at this myself to see if I can isolate the problem

this is inevitable on images where my sort of processing is done, if i get the heavy black lines i, copy a merged image, as this wont work with lots of exposure layers without doing this, then on a fresh empty layer clone the black edges away, then delete the copied layer to leave the cloned layer and all is well