36 + Mp SL?

[CheshireCat]

Advertisement (gone after registration)

So, for half of the 24 million picture elements, you don't know if they were green, but you do know what luminance value the green should be, so you actually have a lot of information about that cell. In addition, you know the pattern of green around the cell based on all the green cells in the area around the cell. This means your software is in a very good position to estimate if the cell should be green or not. Based on this, you will get 24 million elements for each color which all will have the correct luminance value and a very high probability of having the correct color.

 

 

No.

You have absolutely zero information about the luminance of the missing color components.

In other words, for a Red pixel you don't know what the Green and Blue values are.

The demosaic algorithm tries to guess these missing values based on neighboring sensels.

Some of these algorithms just perform plain interpolation, while better ones perform smart guesses and have workarounds to reduce color artifacts. But these missing values are just guesses, as zero information about them has ever been captured.

Of course, if the subject has very low details, even a plain interpolation will work fine.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[adli]

Sorry, but you are wrong. You have 24 million samples of luminance value.

[thighslapper]

The demosaic algorithm tries to guess these missing values based on neighboring sensels.

 

Mr Cat ...... for someone who prides himself on pedantry I am ashamed of you ....... 'estimate' would have been a better choice ...... 

 

Full Definition of guess

1. transitive verb
2.  to form an opinion of from little or no evidence

3.  to arrive at a correct conclusion about by conjecture, chance, or intuition 

[CheshireCat]

Sorry, but you are wrong. You have 24 million samples of luminance value.

 

No you don't, because these "luminance" values are captured after being filtered by a selective-color bayer element (either Red, Green, or Blue).

 

So, for example, take a sample captured behind the Red filter:

- You know what the "luminance" of the Red part of the spectrum is.

- You do not know what the "luminance" of all other spectrum frequencies are.

 

To compute the actual luminance of the sample you would need also a Green and a Blue reading, but you do not have them because they have been filtered out by the Red filter.

 

In other words: whether your subject is red or white (more luminance), that Red sample would have the very same value, so actually, you are not capturing any per-sample absolute luminance value at all.

To solve this problem, demosaic algorithms use the "luminance" values of neighboring Green and Blue samples. This way, 2/3 of the information used to output one pixel is taken from spatially offset samples, hence the loss of spatial resolution.

[CheshireCat]

 

The demosaic algorithm tries to guess these missing values based on neighboring sensels.

 

Mr Cat ...... for someone who prides himself on pedantry I am ashamed of you ....... 'estimate' would have been a better choice ...... 

 

Full Definition of guess

1. transitive verb
2.  to form an opinion of from little or no evidence

3.  to arrive at a correct conclusion about by conjecture, chance, or intuition

 

Let's make it "guesstimate" then, so we are both happy 

[pop]

You are twisting the reality a bit. The M240 has 24 million picture elements capturing light. All of them register the luminance. So you have 24 million samples of luminance values. However, the picture elements can only capture one color each, so you have 12 million capturing green values, and 6 million capturing red and blue respectively.

 

So, for half of the 24 million picture elements, you don't know if they were green, but you do know what luminance value the green should be, so you actually have a lot of information about that cell. In addition, you know the pattern of green around the cell based on all the green cells in the area around the cell. This means your software is in a very good position to estimate if the cell should be green or not. Based on this, you will get 24 million elements for each color which all will have the correct luminance value and a very high probability of having the correct color.

Sorry, this is not correct.

 

For one, none of the sensor sites or cells registers the luminance. The luminance is the total energy that arrives at a given site or point of the sensor. There is a filter in front of each site which discards part of that energy and passes only some of the light.

 

For a site or cell behind a green filter, there is no way of reconstructing how much blue or red light is emitted by the object in that place. All the sensor can tell is that in this particular place the green filter passed a certain amount of light within the bandwidth of the green filter. The same applies to sensor sites behind red and blue filters.

 

In this sense, the sensor sites or cells or elements (called "sensels" by some) are not pixels in the usual accepted sense of the word. Each one carries information about one third of the visible spectrum. A real pixel would carry the complete information about the color, usually expressed in terms of R, G and B brightness, respectively.

 

The conversion of the image from the filtered sited into real pixels with three colors is done by the de-mosaicing part of the raw conversion software. As you point out, there are heuristics which enable the software to supply color values that are missing in the picture as provided by the sensor, but those heuristics are not guaranteed to exactly reconstruct the original scene. Hence, they are not more accurate than a mere interpolation, but they might result in pictures which are more pleasant to look at.

 

There are sensors which do not use any color filter arrays, but these are not what we're discussing here.