MFlenses - Une nouvelle vie à l’ère numérique
MFLenses 2026

Forum des objectifs à mise au point manuelle

Explorez le monde des objectifs anciens à travers des notes de réparation, des photos d’essai réelles et des archives détaillées.

Explorez le monde des objectifs anciens à travers des notes de réparation, des photos d’essai réelles et des archives détaillées.
Pleins feux

Emplacement partenaire mis en avant sur la page d’accueil du forum.

Bayer Filter Covered Sensors
Voir le sujet précédent :: Voir le sujet suivant  


Reply

The BGGR Bayer Filter pattern (image courtesy of wikipedia)



Levels of illumination recorded by sensor covered with BGGR Bayer filter.



Red-filtered photo cells only.



Green-filtered photo cells only.



Blue-filtered photo cells only.



Combine red, green, and blue, no de-mosaic yet.



100% crop. Enlarge to see image is made of red, green, and blue pixels of varying brightness.



After de-mozaic and color balancing.



A silicon photo-cell stores electrical charge when exposed to em energy of any light wavelength; color is not differentiated, response over the visual spectrum is curved, non-linear.

Photo-cells arranged in a grid pattern form a digital camera sensor.

Dense grids of photo-cells are etched into silicon wafers, as for integrated circuits, cpus and such. Sensor array megapixel race requires the best result from silicon wafer manufacturing technology, always pushes the limits of possibility and yield. Canon's recent large format sensor is perhaps current cutting edge example.

A grid of a pattern of red, green, and blue filters, each the size of one photo cell, is aligned over the photo cell grid. There are double the number of green filtered cells, to correspond with greater eye sensitivity to green colors.

Raw data from camera sensor is a list of the levels of charge accumulated by each photo cell. The range of red, green, or blue illumination level at each cell is divided into (today) 512 digital increments (10 bits). Raw data 'development' software constructs an image from the raw red, green, and blue levels of each cell.

First, each red, green, and blue level is multiplied by a certain amount to set white balance, and to compensate for the differing cell response to each color. Color calibration devices seek to match display and printer colors on a test target image with the colors on the actual test target, when viewed under standard lighting.

Next, red green and blue components of each cell are calculated by utilizing the color levels of nearby surrounding cells.
Algorithms for determining individual cell color from surrounding cells vary. Some programs provide several methods. (ref. dcraw)

Non-destructive compression algorithms use various sorting and even reverse-bayer type encoding to achieve smaller file size. Destructive compression algorithms use various color-combination and elmination methods.


Reply

Fascinating, and to me totally incomprehensible.



patrickh


Reply

and where would be the news about it? I'm not getting it...


Reply

Well, I wanted to visualize the histogram data using the image itself instead of a graph. I couldn't find any examples so I wrote a simple program to plot. The images are not the same as for example a red-channel only; here is seen only red filtered photo cell site levels, the other photo-cells are black in these images.

Another sequence, watch the red band on the box Wink:













Reply

I am not sure what the point of this thread is Smile

visualopsins a écrit:

Raw data from camera sensor is a list of the levels of charge accumulated by each photo cell. The range of red, green, or blue illumination level at each cell is divided into (today) 512 digital increments (10 bits). Raw data 'development' software constructs an image from the raw red, green, and blue levels of each cell.


Copy-paste from an old article?


Reply

Anu a écrit:
I am not sure what the point of this thread is Smile

visualopsins a écrit:

Raw data from camera sensor is a list of the levels of charge accumulated by each photo cell. The range of red, green, or blue illumination level at each cell is divided into (today) 512 digital increments (10 bits). Raw data 'development' software constructs an image from the raw red, green, and blue levels of each cell.


Copy-paste from an old article?


Original, and wrong! Embarassed

Now I am confused. With 24-bit color display, R, G, & B each can have only 255 levels. Yet modern camera dynamic range is 12-14 bits, or 4096-16384 levels, and on import into PS the DR is expanded to 16-bit or 65536 levels. Then to properly display all 65536 levels of R, G, and B requires a monitor capable of 48-bit color, right? Is PP software fudging the 48-bit color into 24-bit color for display?


Reply

visualopsins a écrit:
Anu a écrit:
I am not sure what the point of this thread is Smile

visualopsins a écrit:

Raw data from camera sensor is a list of the levels of charge accumulated by each photo cell. The range of red, green, or blue illumination level at each cell is divided into (today) 512 digital increments (10 bits). Raw data 'development' software constructs an image from the raw red, green, and blue levels of each cell.


Copy-paste from an old article?


Original, and wrong! Embarassed

Now I am confused. With 24-bit color display, R, G, & B each can have only 255 levels. Yet modern camera dynamic range is 12-14 bits, or 4096-16384 levels, and on import into PS the DR is expanded to 16-bit or 65536 levels. Then to properly display all 65536 levels of R, G, and B requires a monitor capable of 48-bit color, right? Is PP software fudging the 48-bit color into 24-bit color for display?


In common with Anu, I'm not sure what you are trying to do here. These ideas have been discussed many times in many places. However to address particular points ...

10 bit arithmetic gives 1024 levels not 512.

Citation:
With 24-bit color display, R, G, & B each can have only 255 levels.

There are 256 levels for each channnel - you've forgotten the zero ... 0, 1, 2, ...... 253, 254, 255.

Citation:
Yet modern camera dynamic range is 12-14 bits, or 4096-16384 levels, and on import into PS the DR is expanded to 16-bit or 65536 levels.


Computers tend to work with bytes (1 byte = 8bits) so you can work with 8 bit arithmetic , 16, or 32 or .... If you convert 12 (or 14) bit information to 8 bits you lose information. The dynamic range is reduced and you can get artefacts such as posterisation. You don't lose information with 16 bits. However, Dynamic Range, DR = (largest recordable signal)/(smallest recordable signal). You can't increase it by working with 16 bits.

Citation:
Then to properly display all 65536 levels of R, G, and B requires a monitor capable of 48-bit color, right?


That's 65536 levels for each of the channels ? If so then yes.

Citation:
Is PP software fudging the 48-bit color into 24-bit color for display?

It's probably the video card and/or the monitor which reduces the bit depth. My card tells me that it's 32 bits but I'm not sure if that's 3x10 + 2 bits extra or 3X8 + 8 bits extra. My monitor pretends to be 24 bits but some cheap monitors (like mine !) are really only 18 bit (3 x 6)