3 ms·

Talking about film is particularly complicated, as film does not have an entirely linear response to light. This is called reciprocity failure and means that y

Talking about film is particularly complicated, as film does not have an entirely linear response to light. This is called reciprocity failure and means that you often need to expose way longer than 2x the time to have the effect of 2x the light. For digital, the data directly from camera sensors almost always needs some correction, de-mosaicing or massaging to generate an image viewable on a screen. This requires the camera to make what ends up being an aesthetic decision on what the photo looks like. Detail isn’t just how bright or how dark, but also the available gradients in between. This means there are cases where the dynamic range is automatically expanded (instead of clipped) and contrast unnaturally increased in order to have a photo that isn’t just mud. Ultimately, this means that technical considerations map directly to artistic ones, and there is no objectively correct image from sensor data. The idea that a ‘no filter’ picture conveys some kind of divine truth is a myth.

leephillips 5y ago
Good point about the meaninglessness of a “no filter” concept. The article is very interesting in regard to lightness mapping, and there a another, related subject of hue mapping. This is a good part of the reason why pictures from different camera brands look different. Your description of reciprocity failure is not quite right. The idea is that if you double the time the shutter is open and also decrease the aperture by one stop, you should not change the amount of light hitting the film (you will change other things, of course). The overall brightness should be the same when you make these “reciprocal” adjustments. This does in fact hold pretty well within a certain range of shutter speeds. Reciprocity failure occurs at longer or smaller speeds, where the reciprocal relationship doesn’t quite work.
- antiterra 5y ago
  Night photography, pinhole photography and a number of other common applications absolutely do encounter perceivable reciprocity failure, so there’s nothing wrong with my description.
  - leephillips 5y ago
    That’s fair. I should have offered the additional explanation without saying that you were not right.
jakear 5y ago
Does a “raw” picture convey a divine truth? (Taking “divine truth” here to mean “analog sensor’s signal modulo the specified bit depth and some linear transform”)
- hypertele-Xii 5y ago
  There is no truth. It's lenses all the way down.
- Derbasti 5y ago
  The raw file contains light level recordings affine to the real scene illumination. (The black point is raised, which is why it's affine, not linear). (Assuming no bright pixels are clipped). Each rgb channel is a filtered spectrum. But note that the color filters are essentially arbitrary, and only roughly correspond to the color filters of your eyes or your screen. Also note that some cameras pre-process even raw data, for example with vignetting correction or noise reduction. Some phones even do image fusion and color correction, then re-rawify the data by mosaicing and linearization.
- jillesvangurp 5y ago
  The point is that this is not a linear transform. To make the captured raw data look like a normal photo, a number of steps need to happen and several of those steps are about tricking human brains into believing the photo matches what your eyes would have perceived. These steps aren't actually linear because of a dynamic range mismatch between your eyes, the camera sensor and the output medium (screen and paper). It's a game of fitting the captured scene with e.g. 11 stops of dynamic range to an output medium that has maybe 5 or 6 stops of dynamic range in order to trick human eyes with closer to 20 stops of dynamic range into believing what it sees on the print is "real/plausible". That is of course a very subjective process that involves a lot of choices regarding how to shape your curves and compress the available data. Additionally, human brains do "funny" things with colors and actually re-construct/imagine things that aren't there. Getting the white balance and color grading "right" is another set of steps and what is right depends both on the conditions under which you capture the scene as under which conditions you view the end result. Most phone cameras and in camera processing software algorithms for all of this that produce a certain look that is pleasing under most circumstances. But those are basically just a hard coded interpretation of the raw data. The point of post processing is having finer control over that process and the ability to use more expensive/better algorithms.

leephillips 5y ago

Good point about the meaninglessness of a “no filter” concept. The article is very interesting in regard to lightness mapping, and there a another, related subject of hue mapping. This is a good part of the reason why pictures from different camera brands look different. Your description of reciprocity failure is not quite right. The idea is that if you double the time the shutter is open and also decrease the aperture by one stop, you should not change the amount of light hitting the film (you will change other things, of course). The overall brightness should be the same when you make these “reciprocal” adjustments. This does in fact hold pretty well within a certain range of shutter speeds. Reciprocity failure occurs at longer or smaller speeds, where the reciprocal relationship doesn’t quite work.

antiterra 5y ago
Night photography, pinhole photography and a number of other common applications absolutely do encounter perceivable reciprocity failure, so there’s nothing wrong with my description.
- leephillips 5y ago
  That’s fair. I should have offered the additional explanation without saying that you were not right.

jakear 5y ago

Does a “raw” picture convey a divine truth? (Taking “divine truth” here to mean “analog sensor’s signal modulo the specified bit depth and some linear transform”)

hypertele-Xii 5y ago
There is no truth. It's lenses all the way down.
Derbasti 5y ago
The raw file contains light level recordings affine to the real scene illumination. (The black point is raised, which is why it's affine, not linear). (Assuming no bright pixels are clipped). Each rgb channel is a filtered spectrum. But note that the color filters are essentially arbitrary, and only roughly correspond to the color filters of your eyes or your screen. Also note that some cameras pre-process even raw data, for example with vignetting correction or noise reduction. Some phones even do image fusion and color correction, then re-rawify the data by mosaicing and linearization.
jillesvangurp 5y ago
The point is that this is not a linear transform. To make the captured raw data look like a normal photo, a number of steps need to happen and several of those steps are about tricking human brains into believing the photo matches what your eyes would have perceived. These steps aren't actually linear because of a dynamic range mismatch between your eyes, the camera sensor and the output medium (screen and paper). It's a game of fitting the captured scene with e.g. 11 stops of dynamic range to an output medium that has maybe 5 or 6 stops of dynamic range in order to trick human eyes with closer to 20 stops of dynamic range into believing what it sees on the print is "real/plausible". That is of course a very subjective process that involves a lot of choices regarding how to shape your curves and compress the available data. Additionally, human brains do "funny" things with colors and actually re-construct/imagine things that aren't there. Getting the white balance and color grading "right" is another set of steps and what is right depends both on the conditions under which you capture the scene as under which conditions you view the end result. Most phone cameras and in camera processing software algorithms for all of this that produce a certain look that is pleasing under most circumstances. But those are basically just a hard coded interpretation of the raw data. The point of post processing is having finer control over that process and the ability to use more expensive/better algorithms.