Main page
List of articles

How many pixels does a film frame contain?
Part 2. FAQ


I received a number of responses related to my article about the pixel size of a film frame. Since some of them were truly interesting, I decided to publish my answers as a separate text. I am not a guru who knows the absolute truth. I just want to clarify my point of view. As always, any comments are welcome.

1. Q: "Are you sure pieces of art can be measured by pixels?"

Answer: I did not talk about art. I gave the answer to a specific technical question. It has nothing to do with art. I admit one can create a masterpiece without any pixels at all.

2. Q: "Why is 20 million pixels enough for high quality images?"

Answer: I did not talk about quality either. I derived my estimation based on the equivalent resolution criterion. Quality is a much more difficult thing to analyze and explain. Moreover, it is subjective by nature. Everyone has his own approaches to treat quality. My opinion can be seen from the other answers I gave in this article.

By the way, I think the question "What is better: digital photography or film photography?" is also senseless. Everyone uses what he needs and wants to use.

3. Q: "Why is it necessary to have 3 pixels to resolve a line pair?"

Answer: This fundamental value can be found in many fields of knowledge. As far back as 1933, Kell performed experiments to determine how many television lines it takes to resolve a line pair. He discovered that the total number of TV lines should be multiplied by a factor of 0.7 to produce the resolution. That value is known as the Kell-factor in television. But 0.7 approximately equals 2/3. In other words, it takes three TV lines to resolve a line pair.

The same story exists in the sphere of image publishing. The formula they use looks like this:

(Image resolution in dpi) = 1.5 x (Printing resolution in lpi)

Again, 1.5 = 3/2, and we deal with the same factor.

By the way, two pixels are not enough to resolve a line pair because sensor pixels and lines to be resolved can be arbitrary orientated in relation to each other.

4. Q: "There are many digital cameras with the 50 lp/mm resolution. Thus, film is already dead, isn't it?"

Answer: No, it is still alive. Those sensors with resolving power of 50 lp/mm are smaller than the conventional film frame. We have to enlarge them to a greater degree to get the equal resulting picture. Let us wait until we see a 24 x 36mm sensor with the 50 lp/mm resolution.

5. Q: It looks like you compared the "film plus lens" system and the "sensor without lens" system. Is it correct to do that?

Answer: It is a good question. It is my fault I did not clarified that in my article. Of course, we must compare similar systems, always taking the lenses into account. My conclusion holds for the film and lens against the digital sensor plus the ideal lens. I got the answer on the assumption that the lens will be perfect enough not to spoil the entire resolution of the digital system. Thus, 20 million pixels is a minimum estimation at which digital sensors match film.

6. Serguei F. sent me a message with a lot of critics. I would like to thank him for that. The text was in Russian. I am not going to translate it into English in order not to distort it unintentionally. Still, I would like to place my answers here, since they clarify my position.

A.) The Nyquist theorem and other scientific issues

I am absolutely sure we can do without both the Nyquist theorem and scanning issues when finding the answer to the initial question. Many people do start analyzing scans instead of using their common sense.

Follow my reasoning please:

I shot a roll of film and put it under my microscope without scanning. I saw the resolving power of 50 lp/mm. Then I asked myself: How many pixels would I need to resolve that many lines with a digital sensor, the lens being ideal? I recalled the Kell factor (see Q3) and got the answer — 20 million pixels. It is that simple. I used Occam's Razor to eliminate the Nyquist theorem and other unnecessary factors.

I admit the resolution in the center of the frame is larger. With this in mind, you may want to decrease my estimation of 20 million. On the other hand, three pixels per line pair is a minimum number. Moreover, there is a great difference between the ideal and real lenses. Thus, it looks like you have to increase my estimation to get the true value. Let us assume these factors equalize each other. I suggest we leave the value of 20 million pixels untouched as the first approximation.

B) The theory of information

Many people, including Norman Koren, talk about information when comparing film and digital sensors. That is not a good approach. The amount of information in a photograph is significantly smaller than we can calculate from the sensor size in pixels. Due to this fact, we can compress practically any image by 8 to 10 times in a jpeg file without any visible artifacts. Also due to this, we can use mosaic sensors in digital photography. When we discuss the equivalent size of digital sensors, let us talk about the resolving power instead of talking about information. It is a related but a different story.

C) Film grain

Film grain is not directly connected with the resolving power. The grain may be much bigger than the equivalent pixel size. Look at your magnified film scans to see that.

D) Signal-to-noise ratio (SNR)

Yes, noise is an interesting subject to discuss. Film's grain, quantization, various instabilities produce noise. Yes, SNR does affect the quality. To keep things simple let us not discuss such things here. As far as quality is concerned see Q2.

7. Q: "If we analyze crystals and their properties, we will see that film contains details as small as a wavelength of light. If we take this into consideration, we must count film pixels in Gpixels. Your 20 million is ridiculously small value, isn't it?"

Answer: It depends. I used the criterion of equivalent resolving power. I believe such an approach works in many cases. I think we must always take into account lenses when discussing such things. With optics in mind, you will never approach Gpixels in your estimation.

However, I understand your reasoning quite well. You may say: "Yes, I will not be able to resolve a line pair, but the shape and contours of the spot they produce does matter." If so, then I agree with you. Such a perfectionist like you must develop and print photographs manually. And your works must be masterpieces. You must hate the prints in high quality art magazines as well, since they are below your standards. If not, then I reject your approach. Only one person in a million can have such strict quality requirements. But it is another story.

* * *

Feel free to send me your opinion. It can be published either here or as a separate article. I believe readers must know as many opinions as possible.


Main page
List of articles - Russian Photo Links

Igor Yefremov, 2002, all rights reserved

You must obtain a written permission from me to use any materials of this site for any commercial or non-commercial purposes, unless there is an explicit statement to the contrary.

Hosted by uCoz