I am trying to characterize a camera sensor and I have a problem that I cannot   
   explain. I am trying to measure the linearity of pixels (pix value vs. light).   
      
   My test set up is fairly simple. For a given exposure duration, I measure the   
   FPN and dark noise of the camera (close shutter, dark room, take 200 frames,   
   average out to eliminate temporal noise and what is left FPN and dark noise).   
   I repeat this test    
   for various exposure durations.    
      
   Later, I provide pretty uniform light (via diffuser and point source) and   
   measure pixel values over 200 frames. (no lens) I take out the previous FPN   
   and dark noise average from each of the frame and average out to find out PRNU.   
      
   I repeat the test for various exposure durations under the same light and do a   
   linear regression on pixel values. The linearity of the pixels are very good   
   but they do not cross from zero, the regression line has a DC offset.    
      
   I cannot figure out why? I assumed when I take out the FPN and dark noise   
   components, the DC offset should have been taken out. I use same gain for all   
   measurements. If I repeat the experiment with more intense light, I still see   
   the DC offset on    
   linearity but it is different, somehow the linearity line for pixels is   
   shifted up or down with light, which doesn't make sense.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)