I want to share with you the steps that I am using to avoid flat field overcorrection. This may be useful for people who are using LED panels or T-shirt method and observing overcorrected images.
Here is an integrated image using WBPP where the flat field correction was not proper. I had taken 20 flat images using a LED panel laying on top of the dew shield. The overcorrection is evident on the sides.
When I analyzed the master flat image with a regular sub, I observed that the light fall-off on the flat image is much higher than that in the light frame. As seen below, the illumination drops by 25% (~0.5 to ~0.4) to the edges in the master flat compared to about 10% (~0.064 to ~0.058) drop in the light frame.
A good tool to normalize the master flat in this situation is the FourierTransform tool. I created the magnitude and phase components from the master flat using FourierTransform.
The objective here is to reduce the amplitude difference, so that we get a illumination drop-off comparable to the light frames. I used the HistogramTransformation process to pull down the midtones in the Amplitude component. The phase components shouldn't be touched, as it can alter the image features.
Now a new master flat image can be reconstructed using InverseFourierTransform process using the modified Amplitude and the original Phase components.
With few iterations of trial and error (of fixing the amplitude midtone), a flat field representing appropriate illumination characteristics can be obtained. A test calibration can be quickly tested by the following PixelMath operation.
This is a quick comparison between a previously overcorrected calibrated image and the new calibrated image.
Hope you find this useful.
Thanks,
Anirban
Here is an integrated image using WBPP where the flat field correction was not proper. I had taken 20 flat images using a LED panel laying on top of the dew shield. The overcorrection is evident on the sides.
When I analyzed the master flat image with a regular sub, I observed that the light fall-off on the flat image is much higher than that in the light frame. As seen below, the illumination drops by 25% (~0.5 to ~0.4) to the edges in the master flat compared to about 10% (~0.064 to ~0.058) drop in the light frame.
A good tool to normalize the master flat in this situation is the FourierTransform tool. I created the magnitude and phase components from the master flat using FourierTransform.
The objective here is to reduce the amplitude difference, so that we get a illumination drop-off comparable to the light frames. I used the HistogramTransformation process to pull down the midtones in the Amplitude component. The phase components shouldn't be touched, as it can alter the image features.
Now a new master flat image can be reconstructed using InverseFourierTransform process using the modified Amplitude and the original Phase components.
With few iterations of trial and error (of fixing the amplitude midtone), a flat field representing appropriate illumination characteristics can be obtained. A test calibration can be quickly tested by the following PixelMath operation.
Code:
($T-master_dark)*mean(new_master_flat)/(new_master_flat)
This is a quick comparison between a previously overcorrected calibrated image and the new calibrated image.
Hope you find this useful.
Thanks,
Anirban