An evaluation of image quality metrics aiming to validate long term stability and the performance of NUC methods

摘要

Spatial noise added to temporal noise will affect both the detection and the classification ability of staring image sensors. The spatial noise is due to non-uniform pixels and is also called fixed pattern noise (FPN), though it is not totally static but varies slowly in time, which is due to sensor drift. The sensor drift is mainly due to variability in the ambient temperature and hence the temperature of camera elements, which may be a concern in field trials and the subsequent analysis of the image data. The performance of a non-uniformity correction (NUC) depends on the characteristics of the spatial noise in the image data, in addition to the correction method. In this paper six different quality metrics are studied, aiming to quantify the non-uniformity in collected image data and to validate the performance of a set of NUC methods. The set of methods has been applied on various kinds of real image data recorded with three different imaging sensors in the infrared spectral region, where image data may be severely distorted by fixed pattern noise. Calculated image quality metrics for image data have been compared with results from a visual evaluation. A conclusion is that image quality metrics are useful tools that enable an objective rating of image quality.