Pixels and Subpixels usage in TIFF and RAW

Question

The first answer to this question says that

when saving images in TIFF, each pixel holds three colors' data and because of that, the file is larger... and RAW only hold monochrome data in each pixel and use RGB colored filters and hence the file is smaller

I also didn't get what it's said in the latter sentence. But, what I have learnt is, a pixel consists of three sub-pixels (as in here) (btw, dots means sub-pixels, right?) or four (as in Bayer pattern).
So the my dilemma is, doesn't a pixel hold three colors' data (in sub-pixels) whatever the format is? And did that answer say that the three or four sub-pixels hold a single color (in RAW)?
Please clear me up....

Answer 1

Basically, a RAW file stores data directly from the sensor of your camera. Most DSLR are using what is called a Bayer filter (https://en.wikipedia.org/wiki/Bayer_filter) to retrieve information about color. Usually, for 4 "pixels" (sensitive elements), 2 are used to get information about green, 1 for red and 1 for blue. However, keep in mind that this sensor will provide a 2*2=4 pixels image : a sensitive element on such a sensor isn't a subpixel. I hope the following part will make things clearer :)

In the RAW file of this imaginary 2*2 sensor, you will find 4 values, coded on X bits (12 for example) => your file will weight about 4*X bits.

Because of the Bayer filter, some interpolation has to be done to retrieve color information on all pixels (see demosaicing : https://en.wikipedia.org/wiki/Demosaicing). After this step, each of the 4 pixels will have RGB values, so your file will weight 4*X*3 bits.

As TIFF file are obtained after demosaicing, they are "bigger" : each of the TIFF pixels contains now RGB data instead only R or G or B in the RAW file.

What you are calling subpixel is only related to screen : you need three LED with color R, G and B to recreate a single color (https://en.wikipedia.org/wiki/Additive_color) => a single LED is a subpixel.

You may want to take a look at those questions for more information :

• What is RAW, technically?
• Which is a better way to store your photos? RAW or TIFF?
• What are the pros and cons of different Bayer demosaicing algorithms?

Answer 2

pixel has different meaning in spheres of display technology and sensor technology. pixel of RGB Bayer sensor is represented by only one number (roughly one colour) while pixel of most of RGB graphic displays is represented by three numbers. The number of pixels noted on cameras is not the number of dots divided by four - it is actual number of dots on the sensor which become tristimulus pixels only after making a computational transformation which guesses two additional values for each dot with varying success. This is probably the reason for them to be called pixels, I am not sure.

And, surely, there is a number of file formats to store this information as it is - one number per pixel. The most common one is DNG and may hold recorded sensor data in various styles.

---

There is a display type used by Samsung called "Pentile" which does roughly the same - cuts the chromaticity resolution and somewhat preserves luminance resolution while introducing some colour noise. Those displays have fewer dots than one would expect when learning the nominal resolution. For example, Samsung Wave was told to have 800x480 pixels or whatever Samsung meant to say but upon closer inspection it became obvious that it has as much dots as full colour LCD panels with resolution 400x240 usually have.