Grain and sharpness in scans and enlarger prints

Understanding image sharpness part 8:
Grain and sharpness in scans and enlarger prints
by Norman Koren

Site map/guide to tutorials
Contact | News

Making fine prints in your digital darkroom
Understanding image sharpness and MTF
Image galleries / How to purchase prints
Photographic technique
Image editing with Picture Window Pro

A simplified zone system
Canon FS4000US 4000 dpi scanner
Epson 2450 flatbed scanner
Digital vs. film | Canon EOS-10D
Building simple web pages

updated November 23, 2003

Color Management by X-Rite – Canon – Epson – HP – i1 – Eye-One - Densitometers - Spectrophotometers - Eizo - Canon Large Format

	*Help support this site by linking to these merchants* or		*for books & merchandise*		*for all your photographic needs*
			*for books & merchandise*		*for all your photographic needs*

View image galleries

Table of contents

for the
image sharpness
series

Part 1: Introduction
Part 1A: Film and lenses
Part 2: Scanners and sharpening
4000 vs. 8000 dpi scans
Part 3: Printers and prints
Part 4: Epson 1270 results
Part 5: Lens testing
Part 6: Depth of field and diffraction

Digital cameras vs. film, part 1 | part 2
Part 8: Grain and sharpness: comparisons

Grain introduction | The image | The print
Grain/sharpness comparison | Epson 2450 resolution
Grain aliasing | Software solutions | Nyquist and aliasing

In this page we compare grain and sharpness for three scanners with a well-crafted enlarger print, and we look at grain aliasing and software solutions.

Green is for geeks. Do you get excited by an elegant equation? Were you passionate about your college math classes? Then you're probably a math geek-- a member of a maligned and misunderstood but highly elite fellowship. The text in green is for you. If you're normal or mathematically challenged, you may skip these sections. You'll never know what you missed.

Grain

Grain is a blessing and a curse. Film images are composed of grain, but grain degrades image quality in the eyes of most viewers. As I indicated in Part 1, grain corresponds to noise in a communication system, while MTF corresponds to bandwidth. "Fine-grained" isn't necessarily the same as "sharp." Grain is a random process that can be roughly characterized by two parameters-- average grain size and RMS (root mean square; a common statistical measure) density variation (amplitude). For this reason, there is no single quantitative measure of noise.The visibility of grain depends on viewing magnification. Kodak's Print Grain Index takes this into account. In spatial frequency domain, grain has a broad spectrum with maximum energy between 1/(2*average grain size) and 1/(average grain size). Sharpening boosts noise energy high spatial frequencies, while its inverse, blurring, attenuates it.

Faster films tend to have larger grain-- grain vs. speed is a classic tradeoff. Increasing development time, which increases image contrast, also tends to increase grain size. The choice of developer affects the grain structure-- mostly the RMS density variation. Black & White developers span a spectrum from fine grain developers such as Microdol-X, which softens both grain and image, to ultra-sharp developers such as Rodinal, that have pinprick-fine grain and maximum image sharpness. Sloppy development-- poor temperature control or depleted chemicals-- makes grain uglier: density variation worsens and grains can clump.

Some people prefer sharp grain and choose developers like Rodinal to emphasize it, but most people would be happy to be rid of it. At best they regard it as a necessary evil. Grain tends to be most objectionable in smooth areas like skies.

I wrote this page because I received one complaint too many about grain in the CanoScan FS4000US 4000 dpi scanner. The gist of the complaints was that scanned images looked significantly grainier than camera shop prints; scanner noise and grain aliasing were the suspects. The two most serious complaints involved Kodak Gold 200 film. I decided to investigate by comparing a sharp print with scans from my three scanners: the 4000 dpi CanoScan FS4000US, the 2400 dpi Epson 2450 (to be replaced with the 3200 dpi model 3200), and the 2400 dpi Hewlett-Packard PhotoSmart S20 (recently discontinued). I used a Black & White print because I couldn't find a suitably sharp color print of my own making) I also needed a fairly grainy image to observe grain aliasing, which is rather elusive.

The image

The image I've chosen for the comparison is Swan Window, Phoenixville, Pennsylvania, 1972, part of a series of shop windows that express old-fashioned individuality-- that show no trace of the franchise-driven uniformity that blights our cities and suburbs. It isn't earth-shaking, but it is well suited for the comparison because it's extremely sharp, has a sharp but fine grain structure, and has smooth areas where grain is plainly visible.

It was taken with a Leica M2 with a 1958 vintage chrome-barreled 35mm Summicron lens, at or near its optimum aperture of f/8. A superb lens, even by today's standards. The film was Ilford FP-4, developed in Edwal FG7 diluted 1:15 in a 9% sodium sulfite solution. I used that combination of film and developer because it was extremely sharp, had moderate grain and a nice tonal scale. The grain is tight but has rather high amplitude (density variation) compared to typical color images.

The above image was scanned at 4000 dpi in 16-bit B&W on the Canon FS4000US, boosted in contrast using Picture Window's Brightness Curve transformation (below, right), resized to 400 pixels wide, then saved as a moderately high quality JPEG. The original scan settings were optimized to capture detail over as wide a tonal range as possible. Contrast was boosted to approximately match the print. I didn't take the trouble to match the print exactly because it was evidently dodged (lightened) in the window area and burned (darkened) at the top and bottom; it would have been too time-consuming to do this with the all the scans (no problem doing this for a fine print). The print is pretty snappy-- quite a lot of the shadow detail had to be sacrificed. For reference, a portion of the image with original tones (as scanned) and the Brightness Curve transformation are shown below.

Original contrast (above) before contrast boost (right)

The print

The print was made on a Leitz Focomat 1c enlarger with a 50mm f/2.8 El Nikkor enlarging lens stopped down to f/8 (optimum aperture). Every effort was taken to make it as sharp as possible. The Focomat has anti-Newton ring glass above the film (one side only) to keep it from shifting during the exposure. A well-calibrated Thomas Scoponet (aerial image) focus magnifier was used to verify the image sharpness, and the enlarger had been adjusted so the film carrier, lens, and easel planes were precisely aligned. I was a fanatic printer-- absolutely hardcore-- a trait that hasn't changed. I loved it, and still do, when experienced photographers thought I was using medium format.

The full frame 24x36mm negative was printed 7.6x11.4 inches (194x291 mm; 8.08x magnification) on 11x14 inch Agfa Portriga Rapid glossy paper, selenium toned. The central portion of the print, shown on the right (reduced), was scanned in Black & White at 600 dpi on the Epson 2450 with Unsharp Mask. At 600 dpi, the image is not affected by scanner limitations. At a print magnification of 8.08x, this is equivalent to 4848 dpi on the negative. I resized it to the equivalent of 4000 dpi for comparison with the CanoScan FS4000US, then I sharpened it so the grain and sharpness appear identical to the print viewed through a 10x loupe.

Grain/sharpness comparison

At a monitor resolution of 80 pixels per inch (fairly typical), the entire image would be 70.3x47.1 inches (178x120 cm): almost 50x magnification. The print doesn't look grainy from normal viewing distance, but you can see it clearly under a magnifier.
.

Original print, 8.08x magnification, scanned at 600 dpi and resized with sharpening.	CanoScan FS4000US, 4000 dpi. Scanned with FilmGet, which does moderate sharpening.
Hewlett-Packard PhotoSmart S20, 2400 dpi, sharpened with Unsharp Mask (with threshold), resized	Epson 2450, 2400 dpi, scanned Epson TWAIN 5, which has aggressive Unsharp Masking, resized.

.
Observations

The scans from the print (made in the 1970's) and the 4000 dpi CanoScan FS4000US are remarkably similar. I believe the FS4000US has very slightly more detail, but the difference is hardly significant. I did no sharpening with the CanoScan scan beyond what was done by FilmGet at scan time. The swan's head in the print appears less grainy because it was dodged (lightened).
The 2400 dpi HP S20 is not as sharp as the print or FS4000US, but it's close, as I would expect from simulations. Its grain appears low because the HP scanning software at the default setting does very little sharpening. Unsharp Masking with a threshold to avoid sharpening low contrast regions (i.e., grain) was applied in the image editor.
The 2400 dpi Epson 2450 was less sharp than the HP S20. Using the procedure below, I determined that it is about equivalent to a 1600 dpi dedicated film scanner. The reason is apparently the optical system and/or the CCD sensor, not the diffuse light source, which I discuss in the page on scanners.

Epson 2450 equivalent resolution

The Epson 2450 flatbed scanner scans at the same 2400 dpi "resolution" (actually, scanning density) as the HP S20, but its image has less detail than the S20 and larger grain than any of the other scans. I was able to determine that the 2450 is equivalent to about a 1600 dpi scanner with the following procedure.

I resized the sharp Canon FS4000US scan (upper right in the comparison, above) to the equivalent of 1600 dpi, resized it again to 2400 dpi, then sharpened it so it appears similar to the original Epson 2450 scan. 1600 dpi was closest to the original scan. I couldn't obtain as much detail at 1200 dpi, and there was notably more detail at 2000 dpi.

I must add that the 2450's large apparent grain is not the result of grain aliasing. It is the result of the 2450's low sharpness, which softens both the image and grain, and hence emphasizes low spatial frequencies.

Epson 2450 scanned at 2400 dpi with USM, original size

Epson 2450, 2400 dpi, USM,
original size

Simulated 1600 dpi scan, UnSharp Mask to match 2450

Simulated 1600 dpi scan.
USM to match 2450

Grain aliasing

Grain aliasing is a complex effect that increases the apparent grain size in some digital scans-- it makes individual grains larger and uglier than on comparable darkroom prints. I became aware of it in an article by Pete Andrews, whose defensive tone made it evident that the subject is controversial. The cause of grain aliasing is clear, but identifying it-- distinguishing it from plain unaliased grain-- is not easy. The only way to be certain is to compare a scan with a sharp print of the same image (a 4x6 inch camera shop prints is rarely good enough).

***How to identify grain aliasing***
*Appearance of the scan compared to a well-made darkroom print*		*The cause is...*
Grains are larger with overall rough appearance. Structure is different.		Grain aliasing
Grains are larger but soft in appearance. Structure is similar.		Blurring (misfocus, etc.)
Grains are smaller and sharper. Structure is similar.		The scan is fine. The darkroom print is misfocused.

Grain aliasing appears when two conditions are met.

The scanner must have significant response above the Nyquist frequency f_N, which is defined in the box, below. Since most scanners have little response above the first null, located at 2 f_N, the critical region is between f_Nand 2 f_N. In practice this means the scanner will produce sharp images for its dpi rating.
The grain pattern must have significant energy in the critical region above the Nyquist frequency, between f_Nand 2 f_N. This is not easy to determine by eye. ASA 200 negative film may, for example, have worse grain aliasing than ASA 100 film (which has fine grain that may have low RMS density variation) and ASA 400 film (which has large grain whose spectral peak may be below f_N). Pete indicated that the appearance of grain aliasing is capricious-- hard to predict; it could be affected by development, not just the film type. I concur. I could find no grain aliasing with any of my three scanners in normal operation. To see it, I had to stress the Epson 2450 by scanning a grainy 1974-vintage color negative at 300 dpi.

**Grain aliasing in the Epson 2450 scanner**

1.	2.	3.	4.

is the original medium format image, taken in 1974, when Kodacolor film was a lot grainier than it is today. The portion on the right shows the contrast as scanned. The portion on the left illustrates the extreme contrast boost required to see grain aliasing.
is a portion scanned at 2400 dpi with boosted contrast. What you see is the original grain on steroids (extra contrast).
is a portion scanned at 2400 dpi with boosted contrast, resized to the equivalent of a 300 dpi scan using bicubic interpolation, then sharpened. Bicubic interpolation contains implicit anti-aliasing. No grain is visible.
is a portion scanned at 300 dpi with boosted contrast. Grain aliasing clearly is clearly observable in the sky. It would have been almost invisible without the contrast boost.

Grain aliasing is observable in the 300 dpi scan because the Epson apparently scans every 8th position, without anti-aliasing. This reduces the Nyquist frequency by a factor of 8, but doesn't affect the sensor response; hence there is huge response above Nyquist. Grain aliasing was less obvious in 600 and 1200 dpi scans.

Observations on grain aliasing-- I had a difficult time observing it with any of my scanners, even at 1/4 the maximum optical scan density. The evidence that the CanoScan FS4000US has no grain aliasing is the similar grain structures in scan and the enlarger print. The HP S20 and Epson 2450 both have somewhat larger, softer grain structures, not caused by grain aliasing. It is caused by the response of these scanners, which softens both the grain and the image, emphasizing lower spatial frequencies.

If you haven't worked with fine darkroom equipment or haven't purchased well-made custom prints, you may not realize how grainy film can be (negative film is grainier than slide film). Grain aliasing and scanner noise are often blamed for grainy appearance, when, in fact, straight unaliased film grain is the cause.

Nevertheless, grain aliasing is a real effect worth watching for. I believe Pete saw the real thing: if his 2700 dpi Acer Scanwit 2720 scanner had an optical system as good as the 4000 dpi CanoScan FS4000US, it would have sufficient response above the Nyquist frequency (1350 lp/ inch = 53 lp/mm) to make it susceptible to grain aliasing. Dave Miller of Beckenham, Kent, England, lends support to this theory. Images scanned from color negatives on his 2700 dpi Acer, especially ASA 200, are unacceptably grainy.

Film grain-- summing it up

Film images are composed of grain, and film (especially negative film) can have more grain than most people realize, especially if they haven't seen sharp prints. Film grain is often mistaken for scanner noise or grain aliasing (a real effect, but somewhat rare).

Now the good news. You can reduce grain in software.

Software solutions to grain

The first line of defense is the Blur function found in most image editors. Blur is often combined with a threshold to prevent it from blurring edges. Here is the result of applying Picture Window Pro's Blur transformation to the CanoScan FS4000US 4000 dpi scan (above, right side). Method: was set to Blur More and Threshold: was set to 29.8%. Threshold should be set as low as possible to reduce grain while maintaining sharp edges. The entire image is soft when Threshold: is set to 100%.

Remember, this image is enlarged about 50x on your monitor screen. Very little grain will be visible in a 13x19 inch print (about 13x magnification).

I owe a debt of gratitude to Greg Brakefield (gbrakefield at cfl dot rr dot com, one of the Kodak Gold 200 grain sufferers) for getting me to look more closely at grain reduction software. In October he wrote me complaining about noise in his CanoScan FS4000US scanner. It turned out to be film grain: The structure was identical in scans made with FilmGet (Canon's software) and VueScan. As soon as he realized that returning his scanner wouldn't help, he started testing software. At first he tried demo versions of several Photoshop plug-ins: Grain Surgery ($199), Quantum Mechanic Pro ($199) and Noise Reduction Pro ($99). His favorite was Noise Reduction Pro.

Then he discovered a standalaone program called Neat Image. (Nicole Vincent told me about it at the same time.) An amazing program. It creates a noise profile by analyzing grain in untextured areas of the image, then uses the profile to remove grain. You can control the degree of grain removal and sharpening. Profiles can be saved for use with other images made with same film/scanner combination-- not all images have untextured areas suitable for profiling. The downside: it runs very slowly and there's a learning curve. You may want to start a run before a meal or a coffee break (tea on the other side of the pond). A dual-processor machine makes it easier to perform other work during a run.

You can download a demo version-- very capable, but saves only in JPEG format (not TIFF or BMP), by clicking here.
The basic instructions for getting started are here.

I'll eventually have more to say, but for now, here are Greg's tips:

After the usual Levels and Color adjustments, I ran the corrected image through Neat Image with outstanding results. The following procedure works well with FS4000US scanned images .

Under the "Device noise profile" Tab, pick a detail-free area, and run the "Rough noise analyzer" to analyze it. Then pick several other similar textureless areas and use the "Fine tune analyzer".

Under the "Noise filter settings" tab, I usually leave the "Noise levels" section alone.

Under the "Noise reduction amounts" section, I adjust the "Mid" and "Low" sliders in the to between 90% - 95% (I try keep them equal). I make no adjustment to the "High" slider.

This step is CRITICAL to ensure the images is not over-Filtered to the point that they have a "plastic" look. In the same section (Noise reduction amounts), adjust the "Y" slider to between 40% and 70% (I usually find that 45-60 is best). NOTE: This appears to be the main control for determining how much overall filtration is applied. In addition, adjust the "Cr" and "Cb" sliders (keep them equal) to read 30% HIGHER than the "Y" slider.

Finally, in the "Sharpening settings" section, I've been leaving "Y" checked, adjusting the "High" slider to 60-70%, and adjusting the "Mid" and "Low" sliders to 10 - 20%.

The Key to determining the amounts in step (4) is experimentation. Use your mouse to select an area in the image to analyze, then use the various settings described above. Each time you make a change to a slider, Neat Image will show the changes in the box.
These settings are not etched in stone; they are merely the ones that worked well with my images. If you find better settings, please let me (Greg) know. I'd be happy to send you examples if you like.

If the results in Neat Image's website are any indication, it's well worth the effort. (They've done quite a job on the grainy sail from the middle of Pete Andrews' page.) Here is the result of running Neat Image on an image taken with Kodak Gold 200 film and scanned at 4000 dpi on the CanoScan FS4000US. the crop is magnified 1:1 (1 screen pixel per image pixel). The settings are similar to Greg's. I reduced the Noise reduction amounts somewhat from the defaults: Y = 50% and Cr = Cb = 80%, and I increased the sharpening: 110% for High (spatial frequencies) and 40% for Mid. Impressive results!

Seattle waterfront, Kodak Gold 200, before

Original

Seattle waterfront, Kodak Gold 200, after Neat Image

Processed by Neat Image

The Advanced Sharpen transformation in Picture Window Pro 3.5 does an excellent job of removing grain while optimally sharpening the image. It is described in detail in a PDF manual.

Advanced Sharpen operates in three steps, shown in the dialog boxes below. The histograms represent the differences between neighboring pixel levels-- not the pixel levels, as they normally do.

Step 1 blurs the image. The radius should be set fairly large for effective grain removal. The sliders under the histogram have been set to prevent blurring of high contrast edges.
Step 2 removes black or white specks that result from dust or film defects. There was a problem with small white specks, which were visible after sharpening.
Step 3 sharpens the image with a precisely controllable Radius and Amount. This allows you to find exactly the right value for sharpening without creating "halos" near edges. In the standard Sharpen/Unsharp Mask transformation, you can only set Radius in discrete steps of pixels: 1, 2, 3, etc. The sliders below the histogram prevent low contrast areas from being sharpened. To get the right setting I used quite a bit of trial and error, with the Preview window set to 1:1 magnification.

In Opt (options), I checked Sharpen luminance only, because it gave a slightly more natural result, and I checked High Histogram Expansion. On large files, the Preview refreshes very slowly when you move from Step 1 to 2 or 3 and Blur Amount is greater than 0. On Canon EOS-10D images, where grain is pretty minor, I leave Blur Amount at 0.

The result is excellent: grain removal and sharpness are comparable to Neat Image. But there's one area where Neat Image comes out ahead. It does a better job of maintaining fine low contrast detail, for example, in the railing on the pier above the boat. In general, Neat Image is my tool of choice when the dominant defect is grain. But I regularly use Advanced Sharpen with my Canon EOS-10D. I discuss Picture Window Pro-- my favorite image editor-- here.

Processed by Advanced Sharpen
in Picture Window Pro 3.5

**Picture Window Pro 3.5 Advanced Sharpen dialog boxes, illustrated for each of the three steps.**
Step 1: Blur	Step 2: Speck removal	Step 3: Sharpen

Another Neat Image example-- Dr. P. K. Roy of Durgapur, West Bengal, India sent me this image of his late mother, himself (lower left), and his little brother, taken around 1950. It was scanned from a print made on rough textured paper. I cropped it to save bandwidth. I trained Neat Image on a portion of the background on the upper left, not shown in the crop. It did a near miraculous job of reducing the texture while maintaining important detail.

Original

Processed by Neat Image

Neat Image is an remarkable program for enhancing image quality; one of the most amazing pieces of software I've come across. I added it to my arsenal-- I purchased the professional version-- after several of my friends jumped on the bandwagon.

The Nyquist sampling theorem and aliasing

The Nyquist sampling theorem states that if a signal is sampled at a rate dscan and is strictly band-limited at a cutoff frequency f_C no higher than dscan/2, the original analog signal can be perfectly reconstructed. The frequency f_N = dscan/2 is called the Nyquist frequency. For example, in a digital camera with 5 micron pixel spacing, dscan = 200 pixels per mm or 5080 pixels per inch. Nyquist frequency f_N = 100 line pairs per mm or 2540 line pairs per inch.

The first sensor null (the frequency where a complete cycle of the signal covers one sample, hence must be zero regardless of phase) is twice the Nyquist frequency. The sensor's average response (the average of all sampling phases) at the Nyquist frequency can be quite large.

Signal energy above f_N is aliased-- it appears as artificial low frequency signals in repetitive patterns, typically visible as Moiré patterns. In non-repetitive patterns aliasing appears as jagged diagonal lines-- "the jaggies." Grain, which is random, can appear as large jagged clumps. The figure below illustrates how response above the Nyquist frequency leads to aliasing.

Example of aliasing

Signal (3f_N/2)
Pixels	1	2	3	4	5	6	7	8
Response (f_N/2)

In this simplified example, sensor pixels are shown as alternating white and cyan zones in the middle row. By definition, the Nyquist frequency is 1 cycle in 2 pixels. The signal (top row; 3 cycles in 4 pixels) is 3/2 the Nyquist frequency, but the sensor response (bottom row) is half the Nyquist frequency (1 cycle in 4 pixels)-- the wrong frequency. It is aliased.

The sensor responds to signals above Nyquist-- MTF is nonzero, but because of aliasing, it is not good response.

Many digital camera sensors have anti-aliasing or low-pass filters (the same thing) to reduce response above Nyquist. Anti-aliasing filters blur the image slightly; they reduce resolution somewhat. They're not perfect; sharp cutoff filters don't exist in optics as they do in electronics, so some residual aliasing remains. Lens MTF losses also reduce aliasing.

Images and text copyright © 2000-2013 by Norman Koren.
Norman Koren lives in Boulder, Colorado, founded Imatest LLC in 2004, previously worked on magnetic recording technology. He has been involved with photography since 1964.