Flat-field correction technique for digital detectors

TLDR

Stationary noise patterns and variable pixel responses in digital detectors degrade image quality and reduce detective quantum efficiency, but simple flat‑field image processing can mitigate these effects unless the flat‑field image is taken at low exposure or the system is non‑linear. The study investigates a flexible flat‑field correction that synthesizes a variable flat‑field image via pixel‑by‑pixel least‑squares fitting based on incident exposure. The method uses a linear‑system model that constructs an inverse matrix from a high‑exposure flat‑field image, then applies a pixel‑by‑pixel least‑squares fit storing slope and intercept parameters in two equivalent images to synthesize a variable flat‑field correction. The variable‑exposure flat‑field correction improves detective quantum efficiency, especially at higher spatial frequencies, and has been successfully applied to clinical digital mammography biopsy images, suggesting broader benefit to other digital detectors.

Abstract

The effects of the stationary noise patterns and variable pixel responses that commonly occur with uniform exposure of digital detectors can be effectively reduced by simple 'flat- field' image processing methods. These methods are based upon a linear system response and the acquisition of an image (or images) acquired at a high exposure to create an inverse matrix of values that when applied to an uncorrected image, remove the effects of the stationary noise components. System performance is optimized when the correction image is totally free of statistical variations. However, the stationary noise patterns will not be effectively removed for flat-field images that are acquired at a relatively low exposure or for systems with non-linear response to incident exposure variations. A reduction in image quality occurs with the incomplete removal of the stationary noise patterns, resulting in a loss of detective quantum efficiency of the system. A more flexible approach to the global flat-field correction methodology is investigated using a pixel by pixel least squares fit to 'synthesize' a variable flat-field image based upon the pixel value (incident exposure) of the image to be corrected. All of the information is stored in two 'equivalent images' containing the slope and intercept parameters. The methodology provides an improvement in the detective quantum efficiency (DQE) due to the greater immunity of the stationary noise variation encoded in the slope/intercept parameters calculated on a pixel by pixel basis over a range of incident exposures. When the raw image contains a wide range of incident exposures (e.g., transmission through an object) the variable exposure flat-field correction methodology proposed here provides an improved match to the fixed-point noise superimposed in the uncorrected image, particularly for the higher spatial frequencies in the image as demonstrated by DQE(f) measurements. Successful application to clinical digital mammography biopsy images has been demonstrated, and benefit to other digital detectors appears likely.