Confusion about obtaining linear RGB images.

[vicgalba]

Hi, im working on a color analysis project using Picamera2, and im trying to capture and work with linear RGB images.

I read in the documentation that:

create_still_configuration will use the sYCC colour space by default (by which we mean sRGB primaries and transfer function and full-range BT.601 YCbCr encoding).

Im currently using create_still_configuration() and camera.capture_array() to capture still images, and with this i think i understand that im obtaining sRGB photos. If this is the case, how would you recommend to me to convert them to linearized RGB?

Is there a configuration, maybe, to take RGB photos from the start?

Sorry, i think im going in circles and that this question have been solved multiple times already, but im recently learning color science fundamentals and when it combines with cameras configurations it can get tricky, for me at least.

[davidplowman]

Hi, thanks for the question. Just to understand, are you wanting RGB images without any transfer function applied (therefore linear pixel data), or RGB images with the standard sRGB transfer function?

In the latter case (RGB with the standard sRGB transfer function), just change the output format that you ask for to "RGB888" or "BGR888". There are also 32-bit XRGB versions if you prefer.

In the former case (genuinely linear values), it would probably be best to edit the camera tuning file. You would need to look for the <camera_model>.json (for example, imx708.json), usually found in a folder under /usr/share/libcamera. The precise folder will depend on the model of Pi that you have (there'll be pisp in the path for Pi 5, or vc4 for older Pis). In the file, find rpi.contrast and change it to disable.rpi.contrast. This will turn off the transfer function.

Note that 8-bit output is often a bit narrow for linear pixel data, as you'll find banding might appear if you apply a transfer function later. On a Pi 5 you can get 16-bit outputs, although folks who want to work with linear pixel values often prefer to start with the raw image direct from the sensor. Please do post back with more information on the Pi, camera mode and use case if you would like more advice.

[vicgalba]

Sorry if i wasnt clear. What i want is RGB photos with linear data. Taking them for the beginning with the picamera or maybe transforming (linearize, thats how its called?) the sRGB that im already collecting with the create_still_configuration configuration.

Inside the libcamera i have this multiple files. Which one is the one that i should look for to change rpi.contrast in the case i wanted to turn off the transfer function? Im using the Pi 4.

I would be very grateful if you could give me more advice. As i told you im really new to photography, cameras, and even the usage of Raspberry Pi and Linux itself.

What im doing is some type of instrument for analyzing colour change of samples in time. For this, im controlling the Pi to take multiple photos of a controlled system with a colour checker in it, and then in an analysis process im calibrating (or trying to do it) this photos from linear RGB to Lab. Im somewhat stuck in here, not knowing if working with my sRGB photos is good enough, or even if im doing it right trying to linearize them with different libraries.

From the info i have read i think its fundamental to work with linear data from photos that have not been encoded, and that any type of image processing its done in this type of image and data.

[davidplowman]

Tuning files will be in /usr/share/libcamera/ipa/rpi/pisp (Pi 5) or /usr/share/libcamera/ipa/rpi/vc4 (earlier Pis). You need to locate the file for your sensor (e.g. imx708.json for the regular Camera Module 3, or imx477.json for the HQ camera). If you've built libcamera from source, they'll be in /usr/local/share/.... What camera do you have?

Tip: before changing the tuning file, just type some garbage characters at the very top and check that the camera system doesn't start. This proves you've got the right tuning file and can save significant amounts of head-scratching later!

As I said previously, 8 bit output may not be very good with linear data, but you'll have to judge what works for your application. But do note that there's lots of other processing that happens to the output image - lens shading, colour correction, white balance, colour matrices, denoising etc. which is why I mention that folks making more "scientific" measurements often use the raw image from the sensor.

[vicgalba]

I tuned the file as you told me (i have OV5647) and the images now appear darker.

I guess now they have linear data, even if im still using create_still_configuration configuration, right?

I will look out how to make RAW images,

Thanks for your help and advice!!