The Influence of Data Density on the Consistency of Performance of the Feature Selective Validation (FSV) Technique

Abstract

The human visual system has an immense capacity for compensating for poor or incomplete data. Psycho-visual coding schemes make use of the brain’s ability to extrapolate and interpolate independently of conscious awareness to reduce data (bit) rates but maintain the same level of ‘information’ within a video signal. However, when attempting to produce a simple method for comparing data-sets, primarily for validation of computational electromagnetics, could give rise to a problem. Namely that someone undertaking the visual inspection of (e.g.) modeled data against experimental data will see the same picture whether sampled at N, 100N or 0.01N data points whereas the software undertaking the comparison would process three very different data sets. The Feature Selective Validation (FSV) method was developed to attempt to mimic the group response of a number of experts undertaking the visual comparison. Hence, the quality of performance of the FSV method should not be severely affected by the number of data points if this assertion is to hold, despite the obvious potential for variation. This paper investigates the FSV performance as a function of data density and shows that the accuracy of its performance remains largely unimpeded by variations in the precision of the data supplied.