Exercise 5 - the root of all noise?

Take any picture and convert it to 16 bits per channel.

Starting Cow

Add an adjustment layer and apply a curve to compensate for the camera's sensor linear processing algorithm.


Linear adjustment curve

Save the new darker image and compare it with the original.


Dark Cow

Application of the adjustment curve has resulted in the histogram being skewed significantly to the left hand side.
Now apply a second adjustment curve to return the image to its original state.


Gamma correction curve

This curve removes the skew applied to the histogram and the image looks normal again.

Final Cow

The histogram of the final image is not exactly the same as the original one, this is because the two curves I applied to the image were not exact opposites. 

Looking at the histograms of the images above: -

Ø         Dark cow - in the original image 'seen' by the sensor, the shadow information takes up approximately one fifth of the histogram area

Ø         Final cow - the same shadow information has been 'stretched and moved' and covers approximately two thirds of the histogram area.

Thus, it is hardly surprising that when this linear processing takes place, any noise in the shadow area is 'magnified' as the algorithm guesses to fill in the blanks and make up the new area under the curve.