Must I edit RAW images?

newmikey said:

Raw files ALWAYS reuire editing to turn them into an image of any given format. A raw file is still merely some sensor data and not yet recognizable as a colour image.

Click to expand...

People say this a lot, I'm not sure it is 100% correct unless maybe it's just a matter of terminology.

Somewhere here in a thread sometime back someone finally posted what a RAW file really looked like, and it was mostly crap. The sensor picks up different colors and intensities and records them as numerical values the same as any other image file. So all you need is to take those numerical values, convert them from 14 bit (or whatever) to 8 bit and put them in a JPG. Then you can look at the raw data that the sensor recorded. But apparently the sensor is not linear so when doing a proper conversion to JPG you have to bias the numbers properly, and I seem to remember an issue of one of the channels needing to be doubled or halved. All of that is the job of the RAW converter and its algorithm. If you convert without the algorithm you can get a color image and the subject is recognizable, but the colors in the image will look like crap. So it serves no purpose to view a RAW file, there should always be a somewhat sophisticated RAW conversion involved.

bjn70 said:

newmikey said:

Raw files ALWAYS reuire editing to turn them into an image of any given format. A raw file is still merely some sensor data and not yet recognizable as a colour image.

Click to expand...

People say this a lot, I'm not sure it is 100% correct unless maybe it's just a matter of terminology.

Click to expand...

Good terminology helps you think straight, bad terminology results in muddle. That's why it's best to separate 'processing' and 'editing'

bjn70 said:

Somewhere here in a thread sometime back someone finally posted what a RAW file really looked like, and it was mostly crap. The sensor picks up different colors and intensities and records them as numerical values the same as any other image file. So all you need is to take those numerical values, convert them from 14 bit (or whatever) to 8 bit and put them in a JPG.

Click to expand...

The bit width really doesn't matter very much. You could translate 14-bit raw into a 16-bit TIFF and, if the TIFF was a viewable one, you'd still need to process.

bjn70 said:

Then you can look at the raw data that the sensor recorded. But apparently the sensor is not linear so when doing a proper conversion to JPG you have to bias the numbers properly,

Click to expand...

The reverse. The sensor is linear but the output isn't.

bjn70 said:

and I seem to remember an issue of one of the channels needing to be doubled or halved. All of that is the job of the RAW converter and its algorithm. If you convert without the algorithm you can get a color image and the subject is recognizable, but the colors in the image will look like crap. So it serves no purpose to view a RAW file, there should always be a somewhat sophisticated RAW conversion involved.

Click to expand...

bjn70 said:

newmikey said:

Raw files ALWAYS reuire editing to turn them into an image of any given format. A raw file is still merely some sensor data and not yet recognizable as a colour image.

Click to expand...

People say this a lot, I'm not sure it is 100% correct unless maybe it's just a matter of terminology.

Somewhere here in a thread sometime back someone finally posted what a RAW file really looked like, and it was mostly crap. The sensor picks up different colors and intensities and records them as numerical values the same as any other image file. So all you need is to take those numerical values, convert them from 14 bit (or whatever) to 8 bit and put them in a JPG. Then you can look at the raw data that the sensor recorded. But apparently the sensor is not linear so when doing a proper conversion to JPG you have to bias the numbers properly, and I seem to remember an issue of one of the channels needing to be doubled or halved. All of that is the job of the RAW converter and its algorithm. If you convert without the algorithm you can get a color image and the subject is recognizable, but the colors in the image will look like crap. So it serves no purpose to view a RAW file, there should always be a somewhat sophisticated RAW conversion involved.

Click to expand...

The problem is that the sensor records only a single intensity for each pixel. It does not record color. The sensor itself is a B&W sensor. Color information is captured by placing a color filter in front of each pixel. Half the pixels are behind green filters, and hence measure the intensity of only green light. One quarter or the pixels are behind red filter, and hence those pixels only record data for the red channel. The remaining pixels are behind a blue filter.

If you want to see color, you need to go through a number of steps. The first is to actually figure out the missing two colors for each pixel.

Once you do that you need to normalize the data. You need to account for the color temp of the light, the tone curve, etc.

The data in the raw file does not specify what the final image should look like. There is a lot of ambiguity there, and that needs to be resolved when the raw data is processed. In particular, you need to guess at the missing color channels for each pixel. You also have to backtrack from the imperfect spectral response curves of the colored filters, to estimate what the original color was. That estimation needs to be done in the context of what you think the spectral content of the subject illumination was, and any color cast you think was added by the lens/filter.

The only real issue here is whether you want to use the word "processing" to refer to the many steps between the data in the raw file, and something that actually specifies the colors that should be delayed.

Michael Fryd said:

The problem is that the sensor records only a single intensity for each pixel.

Click to expand...

That isn't a problem.

https://www.dpreview.com/forums/post/66477560

Michael Fryd said:

It does not record color.

Click to expand...

Strictly speaking, nothing records colour. Colour is an individual sensation, like smell

Michael Fryd said:

The sensor itself is a B&W sensor. Color information is captured by placing a color filter in front of each pixel. Half the pixels are behind green filters, and hence measure the intensity of only green light. One quarter or the pixels are behind red filter, and hence those pixels only record data for the red channel. The remaining pixels are behind a blue filter.

If you want to see color, you need to go through a number of steps. The first is to actually figure out the missing two colors for each pixel.

Click to expand...

No, rasterised colours are nothing new. If a viewer for undemosaicked raw files displays red pixel values as red, green pixel values as green, and blue pixel values as blue, you will see colors before demosaicking.

Michael Fryd said:

Once you do that you need to normalize the data. You need to account for the color temp of the light, the tone curve, etc.

Click to expand...

That constitutes a quite ordinary colour management operation.

Michael Fryd said:

The data in the raw file does not specify what the final image should look like.

Click to expand...

Neither does colour negative, even when it is an unmasked film.

Michael Fryd said:

There is a lot of ambiguity there, and that needs to be resolved when the raw data is processed. In particular, you need to guess at the missing color channels for each pixel. You also have to backtrack from the imperfect spectral response curves of the colored filters,

Click to expand...

No recording media is perfect, nor the colour theory is perfect. You can't define "perfect".

Raw is an image in non-colorimetric device (camera, sensor) colour space. Conversion is a conversion from non-colorimetric colour space to a colorimetric one.

--

http://www.libraw.org/

Iliah Borg said:

Michael Fryd said:

...

Click to expand...

Click to expand...

Iliah Borg said:

No, rasterised colours are nothing new. If a viewer for undemosaicked raw files displays red pixel values as red, green pixel values as green, and blue pixel values as blue, you will see colors before demosaicking.

Click to expand...

Are you suggesting that this is what happens in the general case when someone is "viewing" a raw file?

Michael Fryd said:

Iliah Borg said:

Michael Fryd said:

...

Click to expand...

No, rasterised colours are nothing new. If a viewer for undemosaicked raw files displays red pixel values as red, green pixel values as green, and blue pixel values as blue, you will see colors before demosaicking.

Click to expand...

Are you suggesting that this is what happens in the general case when someone is "viewing" a raw file?

Click to expand...

I'm suggesting that your analysis is, to say the least, incomplete, to the point of being misleading.

Implementation details shouldn't be projected onto the nature of raw data, or confused with that nature.

--
http://www.libraw.org/

Iliah Borg said:

Michael Fryd said:

Iliah Borg said:

Michael Fryd said:

...

Click to expand...

No, rasterised colours are nothing new. If a viewer for undemosaicked raw files displays red pixel values as red, green pixel values as green, and blue pixel values as blue, you will see colors before demosaicking.

Click to expand...

Are you suggesting that this is what happens in the general case when someone is "viewing" a raw file?

Click to expand...

I'm suggesting that your analysis is, to say the least, incomplete, to the point of being misleading.

Implementation details shouldn't be projected onto the nature of raw data, or confused with that nature.

Click to expand...

Thank you again for providing corrections so people are not misled.

Far too many people would have simply accused me of being wrong, without attempting to provide correct information. Apparently they are more concerned with claiming that they know more, than sharing that knowledge.

Michael Fryd said:

Far too many people would have simply accused me of being wrong, without attempting to provide correct information.

Click to expand...

I'm not one of those, and it is important to note that I did provide correct information and argumentation. What you do with such information, how you perceive it, is out of my control.

Iliah Borg said:

Michael Fryd said:

Far too many people would have simply accused me of being wrong, without attempting to provide correct information.

Click to expand...

I'm not one of those, and it is important to note that I did provide correct information and argumentation. What you do with such information, how you perceive it, is out of my control.

Click to expand...

Of course.

When you said "No, rasterised colours are nothing new. If a viewer for undemosaicked raw files displays red pixel values as red, green pixel values as green, and blue pixel values as blue, you will see colors before demosaicking."

You were making it clear what happens in the general case when someone views a raw file.

Clearly I was mistaken when I suggested that in the typical case, when people are seeing a reasonable looking color image from a raw file, they are looking at data that has been demosaicked, adjusted for color temp, and had a tone curve applied.

Thanks for making things clear to the beginners in this forum. Far too many people would have jumped to the rare exception, in order to show that they know more than anyone else.

Personally, I think that in a beginner's forum, one is better off describing an overview of the general case, rather than focusing on the exceptions.

I guess my problem is that I have underestimated how many photographers routinely use a viewer for undemosaicked raw files.

My bad.

Michael Fryd said:

Clearly I was mistaken when I suggested that in the typical case, when people are seeing a reasonable looking color image from a raw file, they are looking at data that has been demosaicked, adjusted for color temp, and had a tone curve applied.

Click to expand...

One is mistaken when they deny that raw is an image based on that they usually view raw files after demosaicking etc is applied.

Before one can view a regular JPEG it goes through the reconstruction of missing values, and viewing a JPEG without such reconstruction doesn't result in anything resembling a colour image.

Michael Fryd said:

My bad.

Click to expand...

OK.

Ciao.

--
http://www.libraw.org/