Ember Razement ________________________________ The Fine Art Photography of Malcolm Smith

 

NOISE CONTROL AND SHARPENING

PART A – SHARPENING BACKGROUND THEORY

LINK TO LINK TO
PART B - Sharpening Tools Available
PART C - Some Other Useful Sharpening Tools and Pointers
PART D - Edge Masks
PART E - Real World Sharpening Examples

INTRODUCTION

Early 2007 I bought a copy of the late Bruce Fraser’s book ”Real World Image Sharpening with Adobe Photoshop CS2” which opened my eyes to the sharpening process.  Prior to this I fiddled with the parameters for unsharp mask sharpening (USM) etc. and could not always see a change.  Various authors suggested settings and tried to describe what parameters such as Radius and Amount actually did but there was no underlying theory to explain what was happening.  Bruce outlined much of what is happening and how to control and his book is full of information which I still refer to.  Unfortunately Bruce died and his book leaves many things unfinished – this tutorial starts with Bruce’s ideas and extended by my own thoughts.

I aim to produce fine art prints but you may not be able to spend the extra time on a bulk shoot such as a wedding or product catalog shoots.  Hopefully you can become aware of the problems and if they occur can quickly fix them.

I have carried out a number of experiments and come to conclusions on what, I believe, should be done.  If you think anything I have written is incorrect please e-mail me so we can discuss as I want to lead down useful paths.  The more I ferret out about how things work the more I realize I have yet to learn.

In digital photography there are a number of steps which effect the sharpness of the image and each one has its own characteristics and is best addressed (in Photoshop for example) individually at the appropriate point in the workflow.  I mainly use a high end 21Mp digital camera and wide format ink jet printer for output so my ideas will be aimed in their direction. At the ISO settings I usually work with I don’t often have problems with noise.  Bruce also discusses scanning and other types of printers and noise control so buy his book.

One thing I want you to accept for now is that what most photographers call oversharpening is in reality unfortunate artifacts introduced by the process but which can usually be controlled and eliminated by techniques other than reducing the sharpening Amount.  Digital cameras exceed the sharpness capabilities of film at under 10Mp because of digital sharpening (See Norman Koren http://www.normankoren.com/ tutorials ) so sharpening is important!

Before you start turn off any sharpening in camera – these in camera algorithms are written for processing speed and can be bettered by use of far superior sharpening techniques that can be applied in Photoshop using the techniques described in this tutorial

VISUAL ACUITY

Most sharpening is done by enhancing the acutance (edges in an image - more details later) and it is important that these changes to the image are at just under the limits of visability.

Visual acuity which in layman’s terms is what the viewers eye can just or cannot see.  As this is vital to many aspects of sharpening I want to introduce it here.  Several references Bruces book p60, Roger Clark (Notes on the resolution and Other Details of the Human Eye http://www.clarkvision.com/imagedetail/eye-resolution.html )and others report that experiments on line pairs (a line pair is one light and one dark) show the limit of visibility to be where the angle subtended at the eye of the line pairs is around 1/60 of a degree (1 minute of arc).  This is an average result which depends on the illumination of the target and how good your eyes are but is a useful result.  Simple Trigonometry then shows that the width (of the line pairs) that can be seen at different distances is approximately:

Width := 0.0003 * D     (Width and D same units eg Inches)

Graph of Visual Acuity formula

Line Width Just Seen at Different Viewing Distances

In the function graph above we can see that, as a realistic approximation, the value of 1/100 inch (0.01 inch) is not a bad estimate for viewing prints in a range from 25 inches to 40 inches.  If the print is very big and should be viewed from further away or possibly meant to be viewed closer (such as looking at hand held prints from a portfolio box or in an album) the graph will indicate a new, but not orders different, limit on visibility.  I picture many eyes glazing over with this discussion so in the rest of this document I will simplify the discussion by assuming 1/100 inch is what we are aiming for and ignore viewing distance.

And to those who look at the screen on the back of their DSLR and comment on how sharp the image is all I can say is in their dreams.  I have actually seen and heard these comments being made in all seriousness by professional photographers who should know better!

SHARPENING AREAS TO BE ADDRESSED

There are three areas in digital photography where sharpening is usually required: these are to correct any softening due to the camera (the SOURCE), artistic sharpening to the image (the CONTENT) and sharpening to correct any softening due to the output device (the OUTPUT).

SOURCE Sharpening refers to work required due to blurring in the input device.  This loss of sharpness is a result of a number of factors including demosaicing the sensor image (eg interpolating a red and green colour where there is only a blue pixel sensor), the sensor cells having a finite area (not a point), the effects of the camera anti-aliasing (optical low pass or blur) filter which smoothes out aliasing artifacts, lens diffraction limiting, lens aberration and possibly some out-of-focus and movement blur.

Let us look at the anti-aliasing filter.  My understanding of the Shannon-Nyquist Sampling Theorem is that for anti-aliasing the blur needs to filter out (seriously blur) features sized 2 times the spacing between image sensor cells (or pixels).  This is of course for a perfect low pass filter but a realistic filter would start the blur at wider image objects (and in electronics terms be only so many DB down at the “cut off” frequency).  This is a fairly significant blur and can usually be easily seen in image softness at high magnifications.  As the lens aperture closes (ie larger f stops) diffraction can become larger than the anti alias blur.  For discussion lets assume the source blur is two times the pixel spacing.

I have tried without success to photograph fine edges (ink lines on paper, razor blade edges etc.) to try and measure the camera blur but without any useful results.  However examining some portrait images I have found faint (blurry) single hairs of what looks like one pixel spacing in width which seems to confirms the theory.

There are photographers who claim that (say) a 10Mp sensor is high enough camera resolution to not need source sharpening – is this true?  Basically it depends on the size of the image and the viewing distance – if we limit our discussion to features 1/100 inch we can use the image size command (with resampling disabled) to see how big an image is with two pixel spacings totaling 1/100 inch (or 200ppi).  We get the following table:

SENSOR Mp SIZE (inches) APPROX IMAGE SIZE
11 Mp 20 x 13 Approx A3+
21 Mp 28 x 18 Between A1 and A2

So you may not see the blur on, say, an A4 print.

If you do source sharpening it should be carried out fairly early in the workflow so that with subsequent Photoshop manipulation you are working on a truer representation of the image but do it after any image transformations such as lens or perspective correction (which could nullify some of the sharpening).

Currently for my 11 Mp camera I usually source sharpen for my usual print size of A4 but with my 21Mp camera I only source sharpen for A3 and larger prints.

CONTENTS (The image detail itself) often require sharpening for artistic reasons – lines need to be sharpened to draw attention to them, some lines may of course need to be softened, some areas of texture may need to be emphasized (eg. natural stone texture, the bark on a tree, hair so that the strands can be seen) and again some you don’t want emphasized (eg skin texture or noise in blue skies etc.).  I find that addressing sharpening of lines and textures is best addressed in two stages.  Contrast can also be finely adjusted in localized areas using sharpening tools.

Bruce in his book advances the concept of image detail frequency and offers parameter and other settings to control sharpening and eliminate artifacts.  He defines:

    HIGH FREQUENCY image such as lots of buildings in distance, lots of vegetation such as leaves etc and other images which have masses of fairly close edges
    MEDIUM FREQUENCY images with some sections with fine detail eg cracks in a wood; and
    LOW FREQUENCY images with large areas of low contrast change such as sky and skin but with a few critical edges such as eyes and body outline as in a portrait etc.  Sometimes there are smaller image areas which have different characteristics to the bulk of the image (for example eyes, eyebrows and eye lashes in a portrait) which may have to be addressed separately.

At this stage the image is still in its native resolution and so I usually save it with the word “native” in the file name.

When an image is resized (resampled) to the final size at the required ppi (pixels per inch) some of this content sharpening may be lost but do it anyway.

OUTPUT  Sharpening at this stage is to correct any loss of sharpness due to output device dithering (converting the image to dots on a page) and any ink bleed on the output medium (uncoated paper will bleed more than coated, mat will bleed more than gloss).  Because the softening occurs after the image leaves Photoshop the image on the screen usually must be almost what most photographers would consider over sharpened – unfortunately the only way to test is to print the image.  As in Source sharpening, output sharpening is largely deterministic ie is a function of the printer and therefore the same for all images (but may possibly need to be modified slightly for changes in output paper eg. gloss or matt).

Probably the most appropriate sharpening tool for output is one like Photoshops Unsharp Mask which is an edge enhancing process and in this case the size of the edge enhancing (halos) must be at the limit of visibility – too wide a halo will be seen on the print and a too narrow one not have the desired effect

Earlier on we discussed the limits of visual acuity.  For example when viewing on image at 30 inches an object at the limit of visibility is 30 * 0.0.0003 or 0.009 inches or approximately 1/100 th of an inch (0.01 inch).  So this is a guideline for sharpening halos when viewing a print from 30 inches.

What does this mean in printing terms – well at 300PPI (pixels per inch) which is a commonly used image resolution for sending an image to a printer the halo width needs to be no more than three pixel spacings.  The confusing part here is 300 PPI is the image you send to the printer but the printer may, for example, convert this to 1200 DPI (dots per inch) on the output media but don’t worry – have faith.  For a 300 PPI image to be viewed at normal distances the halos should approach 1/100 inch or in Photoshop terms the Radius of the Unsharp Mask.

An extremely important concept is that the Output Sharpening, and the consequent sharpening halo width, MUST be done at the final image print resolution.  If the image must be resized (resampled up or down) this must be done before any output sharpening is applied.  If the image is to be printed at two or more sizes then resample the native (ie camera) resolution image each time and sharpen each individually.  That is why I wrote under Content Sharpening that I save the almost end result image at the native resolution with the word native in the file name.  So output sharpening is virtually the last step

NOTE:  Always keep in mind the final print resolution.  The notes on visual acuity and what size image component can be seen at the limits of the eyes must be considered.  Output sharpening is vitally important (and should always be done) and so is source and content sharpening for cameras with 12Mp or less but with a 22 Mp or larger camera the source sharpening may not be seen unless a quite large print is being made.

Go to TOPTOP

LINK TO LINK TO NEXT
PART B - Sharpening Tools Available
PART C - Some Other Useful Sharpening Tools and Pointers
PART D - Edge Masks
PART E - Real World Sharpening Examples