Investigation of error factors in analysis of digital noise power spectrum

The noise power spectrum (NPS) measurement is important for assessing noise properties of digital radiography systems, and its measurement method was standardized in International Electrotechnical Commission 62220-1 (IEC). However, improvement of its accuracy is not easy due to random data analysis. In this study, regarding error factors in the NPS measurement using 2-dimensional (2D) Fast Fourier transform, we investigated effects of overlap of region of interests (ROIs), number of average lines in 2D frequency space, directional dependence of frequency property, and detrending techniques. If the number of average lines was set so as to obtain a similar frequency range to IEC, total matrix size was the most important factor and error rate was decreased with increasing of the size. For images, including many trends, detrending using 256 x 256-pixels ROI and second-order polynomial fitting was the most effective. Consistent with the previous report, the overlap of ROIs was not effective for improving accuracy. Contrary to the previous report that indicated effectiveness of 128 x 128-pixels ROI for detrending, we demonstrated less affectivity of the ROI size, other than 256 x 256-pixels.