how to reduce the noises in TEM pictures with imagej

chemaon

Reducing noise in TEM images is a very common task, as these images often suffer from "shot noise" (due to low electron counts) or uneven background illumination. ImageJ provides several excellent tools for this, depending on the type of TEM image you are working with.

Here are the most effective methods to denoise TEM images in ImageJ, ordered from the simplest to the most advanced.

The Median Filter (Best for general noise)
If your image looks grainy (salt-and-pepper noise), the Median filter is usually your best first step. Unlike a Gaussian blur, which softens everything, the Median filter removes noise while preserving the sharp edges of your nanoparticles, cells, or grain boundaries.

How to do it: Go to Process > Filters > Median...
Settings: Set the Radius to a small number (start with 1 or 2 pixels). Check the "Preview" box to see the effect in real-time. If you set the radius too high, you will start losing actual structural details.

FFT Bandpass Filter (Best for HRTEM / Lattice fringes)
If you are doing High-Resolution TEM (HRTEM) and want to see atomic lattice planes more clearly, Fast Fourier Transform (FFT) filtering is the standard method. It separates the periodic signal (the crystal lattice) from the random noise (amorphous carbon or detector noise).

How to do it: Go to Process > FFT > Bandpass Filter...
Settings:
Filter large structures down to: Filters out uneven background illumination (set this to a value larger than your lattice spacing, e.g., 40 pixels).
Filter small structures up to: Filters out high-frequency random noise (set this to a value smaller than your lattice spacing, e.g., 3 pixels).
Suppress stripes: Choose "None" unless you have specific scanning artifacts.

Manual FFT Masking (For isolating specific crystal planes)
If the Bandpass filter isn't precise enough, you can manually select the diffraction spots in reciprocal space to perfectly isolate your crystal lattice.

Go to Process > FFT > FFT. A new window will open showing the diffraction pattern.
Select the Brush Tool or Oval Tool from the ImageJ toolbar.
Draw over the bright diffraction spots (and the central spot) to select them.
Go to Edit > Clear Outside (this turns everything except your spots to black, effectively deleting the noise).
Go to Process > FFT > Inverse FFT. The resulting image will be a beautifully clean reconstruction of your lattice.

Subtract Background (For uneven illumination)
Sometimes the "noise" is actually a gradient where one side of the TEM grid is thicker or darker than the other.

How to do it: Go to Process > Subtract Background...
Settings: Use the Rolling ball radius. The trick is to set this radius to be larger than the largest object of interest in your image. For example, if your nanoparticles are 50 pixels wide, set the radius to 100 pixels. This ensures it only removes the background gradient and doesn't hollow out your particles.
A quick warning on Gaussian Blur
You can use Process > Filters > Gaussian Blur, but use it with extreme caution. It averages the pixels together, which reduces noise but also blurs the edges of your features. It is generally frowned upon for quantitative TEM analysis because it destroys high-resolution data.