Measurement and analisys of spectral noise power in computed tomography

Abstract

Noise is an important feature of image quality. The standard deviation of pixel value in a uniform region has been frequently used as a metric to characterize noise. However, this measure does not provide any information about the noise spatial distribution. A more complete description is given by the Noise Power Spectrum (NPS) which provides both the amount and the spatial correlation of noise. The objective of the present work is to present a methodology and a computing tool to obtain the NPS, in order to analyze its components and study their behaviour for computed tomography (TC) images. Our results show that the major contribution to NPS is a random source for all the explored working conditions. The structural component is constrained to the low frequency region, where it can be as important as the random component. Moreover, we observe that the reconstruction filter and the acquisition technique, axial or helical, have a clear impact on the image noise.