Unable to remove bad rows, bad columns
Added by Michael Cianfrocco almost 9 years ago
Hello,
We are setting up a new installation of Leginon on a Sphera with a US 4000 CCD. While we have been trying to input bad columns and rows to remove when the image is collected the correction node still shows that there are edge pixels present (see the image: No_bad_pixels.png). Confusingly, as soon as we try to remove the edges, we see this edge effect INCREASE, not decrease. See Images: 17_bad_pixels.png and 50_bad_pixels.png.
I'm confusing as to what we are doing wrong - am I missing something?
I should note that we can remove these bad pixels within Digital Micrograph, by specifying which columns and rows to remove.
Thank you for your help!
Mike
| 50_bad_pixels.png (130 KB) 50_bad_pixels.png | 50 bad pixels | ||
| 17_bad_pixels.png (154 KB) 17_bad_pixels.png | 17 bad pixels | ||
| No_bad_pixels.png (154 KB) No_bad_pixels.png | 0 bad pixels |
Replies (2)
RE: Unable to remove bad rows, bad columns - Added by Michael Cianfrocco almost 9 years ago
I forgot to say that we are running myami-3.1 branch revision 19379.
RE: Unable to remove bad rows, bad columns - Added by Anchi Cheng almost 9 years ago
Mike,
You did not do anything wrong, and the program is not broken. What you see with or without bad pixel correction from Leginon is normal.
What you see in the screen shot, whether by Leginon or by Gatan software as seen in no_bad_pixels.png is the result of a typical bad image column/row correction algorithm. When given an image and the columns that are known to be bad, the algorithm do a for loop for each column in the list. It search outward (in this case, away from the edges of the image) until it reaches a column, will call it good column here, that is not in the bad list. The algorithm then reads the intensity of each pixel of the good column, copy them to the bad column at the corresponding pixel position. It then repeat the same on the next bad column. Therefore, you see lines, that would be because there are consecutive columns/rows that are bad since they copy the pixel values from the same good column at its edge.
This algorithm is popular because of its reasonable speed. Because it takes values from the same image, it gives the same statistics property as the image but will be different enough from image to image so that they do not correlate between images, important when doing image analysis.
I can guess where the worse results come from in Leginon correction. If Gatan software gives such a large band of correction when it passes the image as raw image to Leginon, the bright image taken for normalization will also have the pattern, which will cause an extra and unnecessary correction. The effect should cancel out after larger number of averaging when you acquire bright images. You can increase the number of image to average when obtaining Bright image and see if it gets better.
The number of columns and rows Gatan's original is larger than typical. There may be a chance that you will see lines (cross) in your power spectrum. If this is a real problem, a random noise can be added after the data are collected, which I can write a script if you need me to. By the way, if Digital Micrograph bad pixel identification uses a different algorithm that is void of this when it send the image to Leginon, it will be better to specify there. Software at the camera operates at lower level than Leginon, and is usually faster for the same operation.