Abstract
Periodic noise is a common problem in many digital sensors, limiting their performance. The general aim of the research of which this work forms part is to use CMOS sensors in dual energy X-ray mammography. This imaging methodology requires the subtraction of two subsequent images, and it was observed that the resulting image is severely affected by this type of noise. Therefore, methods to reduce periodic noise are required as a preliminary step in the project. This study proposes a software solution to reduce periodic noise, in the form of three cross-shaped notch-reject frequency filters, namely brick wall, Gaussian and interpolation. The suggested filters have been applied during the optical characterisation of a CMOS active pixel sensor, named "Large Area Sensor" (LAS). The optical characterisation methods used were the "photon transfer" and "nonlinear compensation" methods. The former was used to calculate an input parameter for the latter method, which is considered sufficiently precise for a CMOS imager, because it takes into account its inherent nonlinearity. The above filters gave similar results using both the aforementioned methods. Using the latter method, it was found that the periodic noise limited the calculation of the performance parameters of the LAS. More specifically, it leads to an underestimation of the sensor's signal and noise gains by 7% and 13%, respectively. However, its biggest effect is the overestimation of the read noise by 29%. Finally, it underestimated the sensor's full well capacity and dynamic range by 9% and 4 dB, respectively. © 2010 Elsevier B.V.