Flat Field Correction (FFC) in cameras for Machine Vision Part 3
As we have discussed in previous articles, elsewhere on our blog, various types of flat field correction (FFC) exist. Depending on the variant, a flat field correction corrects for dark signal non-uniformities (DSNU), photo response non-uniformities (PRNU) and/or artifacts caused by the illumination and illumination optics. In this blog post we will briefly explain the required steps to make flat field correction work in practice.
Dark and bright field correction
Although the single term flat field correction is used for DSNU and PRNU corrections, in practice correcting for those two phenomena require two different calibration steps.
To calibrate for DSNU a reference image has to be recorded in dark, while for PRNU a reference image has to be recorded with a uniform illumination. These two separate steps in the flat field correction are therefore referred to as dark field calibration and bright field calibration, respectively.
Dark and bright field corrections are thus both part of the flat field correction that corrects for DSNU and PRNU. The dark field correction is the easiest one to calibrate. It only requires a reference image to be recorded without illumination on the image sensor.
The choice of which light intensity to use for the bright field calibration requires a little bit more thought. If you perform the calibration with a light intensity too close to camera saturation you might compensate the camera too much and actually introduce more PRNU for low light intensities. If you use a weak light intensity, the differences in photo response might be too small and you under-compensate the sensor. In general a light intensity that gives a signal somewhere between 40 and 70% of the sensor full scale should give the optimal result.
Low frequency flat field correction
Basically, a low frequency flat field correction is a bright field correction. However as it corrects for low frequency variations which in general are caused by distortions in the optical path and not due to pixel to pixel variations in the photo response, the actual light intensity at which you calibrate the low frequency correction is much less important (in our cameras the average video level has to be between 12.5 and 90%).
Figure 1. The various FFC calibrations have to be performed at a different illumination setting. For dark field calibration no light should reach the sensor. For bright field and low frequency FFC the illumination should be between the indicated borders.
Low frequency flat field correction is thus relatively easy to use. The only thing you have to do is to record a correction image when your reference illumination condition is active and you’re done. There are only a few minor things to be aware of. For example, errors might show up when a sudden peak in intensity is present in the reference scene as the correction is a low frequency correction. Furthermore there might be limits to the total difference that you will be able to correct for. The correction is often achieved by applying a gain per pixel or pixel segment. If the available gain is not sufficient to correct for the difference between the weakest and brightest illuminated pixel segment, a flat field cannot be achieved.
To summarize, depending on the flat field correction variant, reference images have to be recorded in dark and/or in bright field. Although you have to take care of a few things, like making sure the sensor is really dark when performing a dark field calibration and performing a bright field calibration in a light intensity range between 40 and 70% of the sensor full scale. Please contact us if you have any questions.
Links to the other parts: