Ember Razement ________________________________ The Fine Art Photography of Malcolm Smith

 

NOISE CONTROL AND SHARPENING

PART E – Real World Sharpening Examples

LINK TO LINK TO
PART A - Background Theory
PART B - Sharpening Tools Available
PART C - Some Other Useful Sharpening Tools and Pointers
PART D - Edge Masks

This section has some notes on the techniques outlined in Bruce’s book with current thought which will give you a starting point – I hasten to add that you must do your own tests and come to your own conclusions taking note of my previous commentary.

NOTE:  Bruce usually sets layer opacity with sharpening and smoothing down to 65% to 70% so he can increase or decrease the sharpening at a later stage – I started to do this but found I never wanted to change this setting so I never do it.

SOURCE SHARPENING

SOURCE (Digital Camera)
The first step is to address any noise as we don’t want to make this more noticeable by sharpening.  I don’t often need to do this as my camera with the range of ISO’s I use doesn’t result in much noise.  On the odd occasions I do I use Noise Ninja or even simply brush in a blurred layer.

Currently on my 21 Mp camera, unless I intend to do very large prints, I don’t do Source sharpening.  However Bruce’s technique is to:

  • Create a new layer for sharpening.
  • In Layer Style menu (double click on the Layer thumbnail) Set blend Mode to Luminosity, opacity to 65% (to give room for up and down movement) (with next step).  Then set Blend If sliders for “This Layer” to 20/80 and 145/235 to limit the sharpening to the mid tones,
  • Bruce sets the Unsharp Mask parameters to ones he has found useful and which he sets by the camera sensor size (no of pixels and if there is an anti-aliasing filter). He sets Radius as in the table below, Threshold to 0 and Amount to 500 if there is an anti-aliasing filter and 200 if there is none.
Sensor Size in Megapixels Suggested Radius
pixels
LT or EQ 4 1
6 0.6
8 0.5
GT or EQ 11 0.4

Bruce apparently did extensive testing to come to these conclusion (and he is the major author of one of the better commercially available sharpening tools PhotoKit Sharpener) and comments that for very high megapixel cameras such as the Canon EOS1Ds II which is 17 megapixels you could try 0.3 pixel radius but the effect of 0.3pixel radius is “dramatically weaker” than 0.4.  I don’t understand the theory behind this table but perhaps the answer is that very low megapixel sensor arrays have a poorer quality and therefore much more aggressive filters to disguise their lower resolution and higher noise and conversely high megapixel sensors are more expensive and manufacturers can justify a higher quality filter which is less aggressive.  Bruce comments.  Medium format backs are 22 to 60 megapixels but these usually don’t have anti-aliasing filters – I don’t have one or used one and therefore have no experience with them but Bruce suggests a greatly reduced Amount of 200 in the sharpening scheme above.

As we know from previous tests that the difference between Radius of 0.3 and 1 is not the width of the halo but the amplitude.  I suggest using a Radius of 0.7, Threshold 0 and trying amplitude values around 400 (and adjust by lowering opacity down from 100%).

I have been experimenting with sharpening using filters based on deconvolution such as Focus Magic for source sharpening.  My thoughts are that Focus Magic for source sharpening probably should be a 2 pixel diameter to eliminate anti-aliasing blur (from the Nyquist Shannon Sampling Theorem) but this may need to be increased if there is lens (out of focus) blur.  You can use the Focus Magic test probe to try to measure the amount of blur at points of fine detail such as eyelashes.

SOURCE (Scanned Film)

A few notes on Unsharp Mask Source sharpening of Scanned film from Bruce’s book:

  • You may need noise (film grain) reduction on a layer – possibly use a layer mask to protect the edges (don’t invert as for a sharpen mask where you want to protect non edges) – any noise reduction should be done before sharpening.
  • Create a sharpen layer and add a mask to stop non edges being sharpened
  • Run Unsharp Mask Radius 0.8 pixels and Amount 200 - 300

CONTENT SHARPENING

I usually use a global content sharpener of the edges (using an edge mask) followed by a creative sharpen and/or smoothing which I brush in for texture as previously described.  For example in the creative process you may blur background detail to make sure it doesn’t draw eye to an area that is not a focal point of the image or even reduce the sharpness of an edge again to keep the eye away.  This is also the point where I enhance texture and address points like eyes and eyebrows, hair etc.

Bruce introduces the concept of High, Medium and low frequency images.  Sometimes an image may have areas with more than one of the above.  For example eyes, eyebrows and eye lashes in a portrait which may have to be addressed separately.

His technique is with the suggested parameters he uses are”

BRUCE’S GLOBAL EDGE SHARPENING USING EDGE MASKS

IMAGE FREQUENCY LOW MEDIUM HIGH
Create a Sharpen Layer at the top and set Blend If and Opacity to 65% 20/60 & 180/235 20/60 & 180/235 20/60 & 180/235
Do we create a new layer copy of the image which we blur? YES and Gaussian Blur say 10 pixels YES and Gaussian Blur say 1 Pixel NO
Create a new Alpha Channel from Red, Green or use Calculations Yes from the new layer Yes from the new layer Yes from the sharpen layer
To the new  channel Filter>Stylize>Find Edges Yes Yes Yes
Adjust Levels Image>adjustment>levels Yes Yes Not much if any
Invert channel (use <CRRL I> ) Yes Yes Yes
Give the Alpha Channel some more Blur 10 Pixels 0.4 Pixel 0.8 Pixels
Adjust Levels again Yes Yes Not much if any
Make into a selection and add layer mask to sharpen layer Yes Yes Yes
Sharpen R 3 A 200 T 0 R 0.8 A 200 T 0 R 0.6 A 200 T 0

As I have discussed in an earlier section I don’t believe that low frequency images inherently need wider sharpening halos than high frequency images.  I also believe, however, that you should look very critically at all sharpened images looking for problems.  Most of my fine art images are nudes and landscapes and I usually use similar parameters to sharpen them.

What does all this mean?  My current view on sharpen settings is that:

  • Sharpening should be set to give the desired effect at the current resolution to give the appropriate halo width and will be the same for all frequency images.
  • The creation of the mask will be hand tuned for different frequency images.  For example consider a face surrounded by loose hair (hair where you can see the separate strands).  Mostly I use a Radius of 1 or 2.
  • The width of the white mask lines should be sufficiently wide to allow the halos of selected width to be fully seen.
  • I am not sure what the creation of a new layer to blur does.  My current practice is not to do this but select the highest contrast channel in the sharpening channels, copy it and apply blur to it.  My reasoning is that I want full control of the highest contrast channel before blurring distorts things.
  • The above may need to be tuned to individual Images if problems appear.

Don’t forget you can apply two (or more) lots of sharpening to two (or more) layers each with a mask created to target the sharpening of different parts of the image (this will most likely require manual adjustment of the masks with a high opacity black brush).  You can also use the brush in technique for localized areas such as hair or fabric texture (eg denim).

So my process is currently:

  • Create a Content sharpen layer and use layer Style to set Blend If sliders to 20/40 & 190/235, Blend Mode to Luminosity and Opacity to 100%.
  • Open the Channels palette:
      • Create a new alpha channel by copying one of the existing (eg R or G) colour channels or combine two channels with Calculations and Pin Light Blend to form the best highest contrast alpha channel for constructing the mask.
      • Blur the new mask channel
      • Run Filter> Stylize> Find edges on this channel
      • Adjust with Levels
      • Invert
      • Adjust if necessary with more Gaussian blur to kill any mid tone posterisation or noise and more Levels adjustment.
      • Load this mask as a selection and add the mask to the sharpen layer.
      • You can delete the alpha channel used for creating the mask (but not the mask itself).
  • Sharpen using Unsharp Mask.

Some final notes.  When I adjust with Levels the first time (before Inverting) I always adjust the white point to where the tail of the histogram hits the baseline and sometimes I tweek the gamma (mid pointer) a bit to reduce some of the posterisation or low level unwanted texture but not the black point.  After I invert the mask to get the white lines where I want sharpening to be seen and further blur (to further address posterisation) I again in Levels move the white pointer to where the Histogram hits the baseline and then tweek (fiddle with) both the black and gamma pointers.  Don’t forget you can paint over bits and you can also try strengthening the paler gray lines with the Dodge tool set to adjust (Range) highlights.  You can also go back to the Mask and make some further adjustments after sharpening (eg paint out bits, give further blur, Levels etc).

OUTPUT SHARPENING

My notes will be mainly limited to Ink Jet printer sharpening as this is how I produce my fine art prints.  Bruce’s book considers the wider application including continuous tone printers and traditional ink (offset) printers.

Output sharpening for Ink Jet Printers is to combat the effects of softening caused by both printer output dithering (as programmed into the printer by the manufacturer and usually held fairly closely by them) and possible bleed in the ink soaking into the paper (this happens more on matt paper than gloss/semi gloss papers).

All output sharpening needs to be applied last with the image at the actual output device resolution (resampled)  with sharpening settings tuned to this (ie output sharpening doesn’t usually depend on image content etc.).  The effect obviously can’t be seen properly on a monitor screen as the blur (dithering, ink spread etc.) occurs when the image is well past the screen display and in the printer.  Therefore the process needs to be largely tuned by trial and error with the image possibly even having the appearance of oversharpening crunchiness on the screen.  Printing on the desired paper (Matt paper usually requires more sharpening than full and semi gloss paper etc.) is really the only way of seeing if the sharpening is appropriate.

I generate my prints on my 12 ink Canon IPF5000 (a wide format hi-tech printer which handles up to A2 paper) at 300 or 600 ppi (the printer is 1200 DPI) so to get sharpening halos at 300ppi or 600 ppi of 1/100 inch the desired halo is around three or six pixels wide.  If necessary protect the highlights and shadows with Blend If sliders in the Layer Style dialog box.

Bruce proposes a rather neat ink jet sharpening using Unsharp Mask followed by the High Pass filter.  For a 300 ppi inkjet print on glossy or semi glossy paper he:

  • First resamples to the desired output resolution with  BICUBIC SHARPER to reduce and  BICUBIC SMOOTHER to enlarge.
  • Create a new sharpen layer at the top of the stack with Normal Blend mode
  • Set Blend if sliders to 10/20 and 230/250
  • Sharpen with Unsharp Mask with Amount 320, Radius 0.6 and threshold 4.
  • Immediately EDIT>FADE and in the Fade dialog box set Opacity to 70% and Blend mode to Luminosity
  • Change Blend Mode of layer (in the Layers Palette) to Hard Light with 50% opacity
  • Run the High Pass filter with a radius setting of 2 pixels

Essentially I understand that this enhances the edges of the image before applying the High Pass to this enhanced image.

With my Canon IPF5000 printer I don’t use the Photoshop print command as Canon have provided an export interface which sends a greater bit depth to the printer than the eight bits of the print command..  It prints at 1200 dpi but the image can be sent to the printer at 300ppi and 600ppi.  Mostly I send prints at 300ppi but I have heard of tests being done where the difference between 300 and 600 could just be just seen.  Often if the image doesn’t have to be massively resampled I send it to the printer at whatever ppi it is in and let the printer resample and it does a very good job of this – I do this for most of my test/proof prints but I always resample my fine art customer prints.  The printer can also be instructed to print with the head moving in one direction only (unidirectional printing) to achieve greater accuracy in placing the ink dots.  My printer, like my monitors, is profiled regularly with a Greytag Macbeth Eye One photometer.

Output to screens does not have the blurring and dithering of ink jet printers and so requires less sharpening.  Screens are usually spoken of as being 72dpi but on three of my modern screens (a 17inch HP, a 13.3 inch Sony and my Eizo 24 inch screen) the resolution is about 100dpi or slightly above.  This means that the halos we should be aiming for are one pixel spacing wide. (of course we know that we can’t get narrower than this).  My current settings for on screen and web sharpening with Bruce’s technique are:

  • Create a new sharpen layer at the top of the stack with Normal blend mode
  • Set Blend if sliders to 10/20 and 230/250
  • Sharpen with Unsharp Mask with Amount 150, Radius 0.7 and threshold 4.
  • Immediately EDIT>FADE and in the dialog box set Opacity to 50% and Blend mode to Luminosity
  • Change Blend Mode of layer to Hard Light with 30% opacity
  • Run the High Pass filter with a radius setting of 0.7 pixel

All of these suggestions are listed for you to use as a starting point and, particularly with the knowledge of how halo width changes with Radius, you should do your own tests and come to your own conclusions.  Always look at the results on your final output medium (eg ink jet print) and don’t get waylaid by what is on the screen, particularly when the image on the screen has been zoomed in to a large magnification.

Finally do your own tests, ask the right questions and come to your own conclusions on appropriate settings.

SUMMARY

Let us summarise some of the important points.

  • Over sharpening is usually caused by halos being too obvious (too wide) and clipping to white and usually not the amount of sharpening..
  • Visual acuity investigations have given guidelines for what can just be seen and sharpening halos should be just below this point.
  • Sharpening halos are discrete and in multiples of one pixel spacing wide (ie there can be no halos of 0.5 pixel – it will be 1 pixel spacing)
  • The intensity of halos for low Radius decreases as the Radius decreases.
  • The intensity of the halos usually increases as the difference in original image change increases.
  • Halos that clip (or near clip) to white) are more obvious and need to be toned down usually with Blend If sliders in Layer Style dialog box.
  • Sharpening is needed to correct input device softening (the SOURCE), output device softening (OUTPUT) and creative sharpening of the actual image contents (CONTENTS).  Source and Output are largely deterministic (depends on the input and output devices).
  • Wider sharpening halos don’t obscure fine detail.
  • Edges should be sharpened not wide areas of relatively flat tone.  Noise in skies and skin texture should be protected from sharpening.
  • The way edges are sharpened and wide areas protected is to use edge masks.

CASE STUDIES

Case Studies is a later addition to my notes and is intended to illustrate problems I have encountered and how they were solved.  I intend to add cases as they occur from time to time.

CASE 1 - PREGNANT NUDE

Prehnant Nude
In this image of a pregnant nude wearing a straw hat there was only one area of potential problem and that is the halo around the dark breast against the darkish background.  The problem is twofold: first the halos may be seen and secondly we need to decide if the area is to be sharpened at all (as it will attract the eye from possibly the main point of the image).  The problem is that (using a Threshold test layer) the brightness of the halo is a maximum of 150 (in the normal range 0 black to 255 white) and is way below the point where Blend If sliders can be used.  The solution of course is to work on the edge mask with a low opacity black brush in this area – mild painting in reduces the intensity and further work will hide the sharpening halos completely.

Some areas of the straw hat clipped but were easily controlled with Blend If sliders.  The halos discussed above on the dark breast possibly would not be seen on the final print but the sharpening would.


Go to TOPTOP
LINK TO LINK TO
PART A - Background Theory
PART B - Sharpening Tools Available
PART C - Some Other Useful Sharpening Tools and Pointers
PART D - Edge Masks