Computer Science, Engineering and Systems Publications List Information Warfare, Hypergames, Systems Research Ad Hoc Networking Research Computer Architecture Research - Password Capability Systems Industry Publications Industry Hardware Design Projects Interesting Papers Photo Galleries Biography Email Carlo GOTO Home

Last Updated: Mon Aug 4 05:36:56 UTC 2014

Some Observations on Digital vs Wet Film Photography

Dr Carlo Kopp, MIEEE, SMAIAA, PEng
Text, Images 2010, Carlo Kopp



A question which is frequently asked is whether digital cameras produce  “better”  images than film cameras and scanners. This is a somewhat vexed question, given the enormous differences between various digital and film cameras, in terms of lens performance, CCD/CMOS array noise versus film emulsion grain performance, anti-aliasing and demosaicking algorithm performance, and specific lighting conditions on the day.

The usual argument then often distills down “what number of Megapixels in a scan of a film frame is equivalent to what number of Megapixels in a digital camera image?”; assuming of course all else being “equal”.

This question is more easily answered, as it is a straightforward application of Nyquist's sampling theorem.

Most digital cameras employ Bayer pattern technology to separate red, green and blue channels.  The  Bayer pattern is a clever way of cheating on imaging array complexity, by exploiting the limitations of the typical human eye, which can resolve objects more sharply in monochrome, compared to the red, green and blue bands, yet the eye is typically most colour sensitive in the green band.

If we assume that the spacing between pixels is geometrically equal, and of value a, then in a typical linear CCD scanner used to digitise slides or negatives, each pixel is imaged in the red, green, and blue bands, and the highest spatial frequency in the image which can be captured without aliasing is exactly one half of the spatial frequency determined by the spacing between pixels, a.

Let us now consider the Bayer pattern CCD/CMOS chip. If this chip has an equal spacing between pixels, then if it images in the green band, the shortest distance between pixels is 1.414 x or sqrt(2) x a. This sets a hard limit in turn on the spatial frequency the chip can image, to one half of the spatial frequency determined by the spacing between pixels, 1.414 x a. Applying the same argument to the red and blue channels, yields 2 x a.

What does this mean in terms of
Megapixels? When people describe imaging performance in Megapixels, it is a measure of sampling density across an area, so assuming equal  or similar aspect ratios in image frames, we simply square the representative ratios. The aim is to find how many more Megapixels does a Bayer pattern CCD/CMOS chip need compared to a film scanner, to achieve the same spatial frequency and thus image sharpness. The same argument, incidently, can also be applied to comparisons between imaging site (pixel) density in the Foveon technology CCDs.

The green channel determines the ultimate limits in image sharpness for the Bayer pattern
CCD/CMOS chip, which must have in this ideal case twice as many Megapixels compared to a good quality scan of a wet film frame. If the scene being imaged has large areas of finely textured red or blue colour, then resolution in the red or blue bands becomes a quality factor. In both the red and blue channels, the Bayer pattern CCD/CMOS chip needs four times the number of Megapixels required by the scanner.

This limitation of the Bayer pattern is typically prominent where the image contains large areas of highly saturated blue or red. Where all red or blue areas in the image have low saturation, or areas have other colours with substantial green spectral content, then software in the camera can be employed to interpolate, and use all three colours to achieve nominal Megapixel resolution.

Four sample images of PC-9/A aircraft at the bottom of this page provide a good example. The images produced by the digital cameras are very good, but lack the fine texture in the smoothly curved and polished red areas which are well captured by the Noritsu HS-1800 scanner off the 120 format film. The fifth and sixth images of cars, scanned by a Fuji Frontier SP-3000, show similar behaviour. This effect is a direct result of the fine surface texture and spectral composition of the reflection.

For comparison, the two images of foliage, dominated by green and yellow tones, are exceptionally sharp and well textured. Put simply, the Bayer pattern likes images with strong green content.

Prima facie this suggests that it is not yet time to throw away your wet film camera, especially if it is a high quality medium format design. But theoretical imaging resolution is not the whole story.

Modern digital cameras have often sophisticated autofocus mechanisms, and often very smart aperture/shutter control algorithms. Moreover, good DSLRs or superzooms have multiple image stabilisation mechanisms to compensate for fine jitter. Many digital cameras include sharpening algorithms. These features are typically absent in film cameras, and the result are often  photographer-induced defects in exposure, field of focus and fine jitter.

Film grain is another impediment to high quality in wet film images. Many films which on paper have respectably low RMS grain size are disappointing in practice. While the RMS grain value may be low, the statistical properties of the grain may still be such, that grain will be present at sizes comparable to a pixel and impair the image.

While noise produces a similar impediment in CCD/CMOS imaging chips, it is generally less prominent than film grain, and also can be better managed by the design of the imaging sensor.

A factor often ignored in marketing literature is the physical size of the pixels in a CCD or CMOS imaging chip. At a constant chip size, increasing the number of pixels reduces the area of each pixel, in turn reducing the number of photons it can capture. Less photons captured in turn means lower Signal / Noise Ratio (SNR) and thus noisier image. Typically lenses are designed for some imaging area, be it chip or film, and this sets contraints, as does the manufacturing yield of the imaging chip. Excessive noise can degrade a digital image just as effectively as grain can degrade a film image.


In conclusion, digital imaging technology using CCD or CMOS imagers will outperform scanned film, but only if the imaging device has a sufficent pixel count.



Example 39 Megapixel image (Joseph Holmes, 2009) produced with a Phase One P45+ digital back, Mamiya 645AFDII, 55 -110 mm f4.5 lens at 80 mm, ISO 50, 1/60: Crater Lake 39 MP

Emil Martinec, Noise, Dynamic Range and Bit Depth in Digital SLRs, Enrico Fermi Institute, University of Chicago.

Camera Sensor rankings with DxOMark




135 / 35 mm Slide Scan Resolutions
36 mm x 24 mm = 1.417 in. x 0.945 in.

PPI
Width
Height
Line Scan Resolution [MP]
Bayer Green [MP]
Bayer Red/Blue [MP] Pixel [μm]
1200
1700
1134
1.9
3.9
7.8
21.1
1700
2400
1600
3.8
7.7
15.4
14.9
2000
2649
1776
4.7
9.4
18.8
12.7
2170
3075
2027
6.2
12.4
24.8
11.7
2400
3400
2268
7.7
15.4
30.8
10.6
2550 3613
2409
8.7
17.4
34.8
10.0
2730
3873
2581
10.0
20.0
40.0
9.3
3000
4251
2835
12.1
24.2
48.4
8.5
3200
4534
3024
13.7
27.4
54.8
7.9
3600
5101
3402
17.3
34.6
69.2
7.1
4000
5668
3780
21.4
42.8
85.6
6.4
6000
8502
5670
48.2
96.4
192.8
4.2


120 / 220 / 6 x 4.5 cm Slide Scan Resolutions
56 mm x 45 mm = 2.36 in. x 1.77 in.

PPI
Width Height Line Scan Resolution [MP] Bayer Green [MP] Bayer Red/Blue [MP] Pixel [μm]
600
1416
1062
1.5
3.0
6.0
42.3
700
1652
1239
2.1
4.2
8.4
36.3
800
1888
1416
2.7
5.3
10.6
31.8
1185
2796
2048
5.7
11.4
22.8
21.4
1200
2832
2124
6.0
12.0
24.0
21.1
1300
3068
2301
7.1
14.2
28.4
19.5
1460
3441
2539
8.7
17.4
34.8
17.4
1550
3658
2744
10.0
20.0
40.0
16.4
1700
4012
3009
12.1
24.2
48.2
14.9
2000
4720
3540
16.7
33.4
66.8
12.7
2400
5664
4248
24.1
48.2
96.4
10.6
3000 7080
5310
37.6
75.2
150.4
8.5
3077
7264
5440
39.5
79.0
158.0
8.3
3600
8496
6372
54.1
108.2
216.4
7.1
4000
9440
7080
66.8
133.6
267.2
6.4
6000 14160
10620
150.4
300.8
601.6
4.2


Representative Film Resolving Power Examples [1985 - 2010]

Wet Film Type
Chart Contrast 1.6:1
Chart Contrast 1000:1 RMS Grain
Notes
Fujifilm Velvia 50 [RVP50]
80 [lines/mm]
160 [lines/mm]
9 Reversal / 2010
Fujifilm Provia 100F [RDPIII]
60 [lines/mm] 140 [lines/mm] 8 Reversal / 2010
Fujifilm Astia 100F [RAPF] 60 [lines/mm] 140 [lines/mm] 7 Reversal / 2010
Fujifilm Pro 160C/S
63 [lines/mm] 125 [lines/mm] 3
2010
Fujifilm Pro 400H
50 [lines/mm] 125 [lines/mm] 4 2010
Fujifilm Pro 800Z
50 [lines/mm] 115 [lines/mm] 5
2010
Ektar Pro 100 [Ekt 100]
-
154 [lines/mm] Very Low
2010
Ektachrome 100G/GX 63 [lines/mm] 140 [lines/mm] 8 Reversal / 2010
Ektachrome 64 [ER]
50 [lines/mm] 125 [lines/mm] 11 Reversal / 1985
Kodachrome 25 [KM] 63 [lines/mm] 100 [lines/mm] 9 - 11
Reversal / 1985
Sources:Fujifilm, Kodak datasheets.



These normalised MTF curves were produced by sampling MTF curves in Kodak and Fujifilm datasheets, and producing a smoothed plot of these sample points, normalised to the peak response. As the curves are shallow at the point of intersection with the 50% MTF line, caution should be exercised in interpreting the curves - they should be treated as indicative rather than absolute. As neither manufacturer has disclosed tolerancing data for their published MTF curves, actual performance may vary and where the plotted film sharpness is very close, the relative rankings might not be reflected in actual film performance. Empirical observation suggests some of the published curves may be pessimistic, especially for the older film types.


Representative Slide and Print Film RMS Grain
Sampling Strategies:
Strategy A: Nyquist (pixel width   0.5 ∗ RMS Grain)
Strategy B: Pixel as integrator (pixel width   2 * RMS Grain)

Film Medium
RMS Grain Size
Strategy A [PPI]
Strategy B [PPI] Notes
Kodachrome 25
9 - 11
4800
1200 S
Ektachrome 64
11
4800 1200 S
Fuji Provia 100F
8
6400
1600
S
Fuji NPC 160 4 12800
3200
P
Fuji Pro 160S/C
3
16000
4000
P / C
Ektar Pro 100
<3
16000
4000 P


Representative Scanner Resolution Performance


Scanner Type
120/220
135
Resolution Limit [PPI]
Colour Resolution [bits]
Notes
Kodak IQSmart/EverSmart Y Y 5,600 16.0 RGB
Kodak HR500 Y Y 5,600 (CCD 6,000)
-

Nikon Coolscan LS-5000ED
N
Y
4,000
14.0 RGB

Nikon Coolscan LS-8000ED Y Y 4,000 14.0 RGB
Nikon Coolscan LS-9000ED Y Y 4,000 16.0 RGB
Noritsu HS-1800
Y Y 2,000
-

Fuji Frontier SP-3000
Y Y 1,700/120, 3850/135
-



Representative Imagery Examples


PC-9/A aircraft scanned to 1200 PPI / 5.8 Megapixels from a 120 / 645 frame of Fuji SHGII 100 film,  cropped and reduced to 768 pixel width. Note the prop spinner (Mamiya 645/1000S).





PC-9/A aircraft imaged by Bayer pattern CCD digital camera at 5.1 Megapixels (below) and 8.0 Megapixels (above), and reduced to 768 pixel width (Fuji S5600 and S5800).





Ford F-100 Pickup, 1953, and Ford Mustang Fastback, 1966, shot with a Mamiya 645/1000S using Fuji Provia 100F and NPC 160, and scanned using a Fuji Frontier SP-3000.





Banksia integrifolia or Coast Banksia (below) and Eucalyptus caesia or Silver Princess (above) imaged using a Fuji S5800 and HS10 respectively. Note the lack of sharpness in the red eucalypt blossoms, compared to the leaves.






Computer Science, Engineering and Systems Publications List Information Warfare, Hypergames, Systems Research Ad Hoc Networking Research Computer Architecture Research - Password Capability Systems Industry Publications Industry Hardware Design Projects Interesting Papers Photo Galleries Biography Email Carlo GOTO Home
Artwork and text 1994 - 2010 Carlo Kopp; All rights reserved.
$Revision: 2.276 $