Friday, January 31, 2014

Samsung Proposes Way to Compensate the Local Black Level Variations

Samsung patent application US20140014818 "Pixel array, image sensor including the same, and method of compensating for local dark current" by Young Sung Cho, Dong Jae Lee, Tae Chan Kim, and Tomer Livneh propose a simple approach to compensate the black level variations across the pixel array. The idea is simple: a number of dark, light-shielded pixels is distributed in random or not-so-random way across the array, and their interpolated output is used for local black level calculations:

There are few modifications that need to be added to that application:

  • Light leaks or photoelectrons diffusion to the shielded pixels can distort the black level measurements. Unless this is completely resolved in the company's ISOCELL technology, one might need to light-block the larger groups of pixels.
  • dark current varies a lot on pixel-by-pixel basis. So, it needs to be somehow averaged across the neighboring dark pixels
  • there are hot pixels, whose dark current is just abnormally high and carries no information about other pixels in the area. They need to be excluded from the dark current calculation process.


  1. How much of the variation of the dark current is of high spatial frequency (i.e. pixel to pixel) and how much of it is of low spatial frequency (i.e. pixel neighborhood to pixel neighborhood)? The method proposed only addresses the latter.

    1. For 4T pixel, the fast spatial pixel-to-pixel DC variations have stddev of same same order as the average DC, give or take.

      Temperature and slow process gradients are the major factors behind the low spatial frequency DC variations. Temperature gradients depend on the sensor power, chip floorplan, and its package and camera module design. I'm not familiar with Samsung sensors and cameras, but for a generic mobile sensor it's not uncommon to have temperature gradients of 2-5C. This can cause a change of the local DC of about 20-70%.

      With these numbers, the fast spatial DC component is probably bigger than the slow gradients, thus the spatial averaging is needed. I guess Samsung is using that, just not writing in the patent.

  2. Check out this one : JP 2007019820A submitted by Fujifilm.
    In my opinion, this is exactly the same as the Samsung application.

    1. Yes, indeed, looks quite similar. It appears that Fujifilm abandoned this application. It has been filed in 2005, published in 2007 and has no other record since then.


All comments are moderated to avoid spam.