Nonlinear Processing Of Quantitative Thermographic Images

Publisher Website
Google Scholar
Abstract
The minimum resolvable temperature difference in quantitative thermographic imaging is essentially determined by photo-detector noise processes. In fast scan devices the overall detector noise envelope may be ±0.5°C. When imaging quasistatic thermal scenes, integration of successive frames (signal averaging) can be used to reduce the effect of detector noise. However, many interesting thermal scenes are rapidly changing, such as the thermal pattern resulting from a laser pulse or from radio frequency energy applied to tissues, and frame integration cannot be applied. In the transient temperature measurement case, some smoothing operator must be used to improve the minimum resolvable temperature difference. Linear smoothing operators, such as averaging and gaussian filters, all have the undesirable side effect of smearing the few edges and other small details in the thermal image. Median filters offer the potential of filtering detector noise while preserving edges and other details. Because the median operator is inherently nonlinear, its effect on the accuracy of measured temperatures cannot be predicted a priori. For the photodetector studied, photoconductive mercury-cadmium-telluride, the detector noise process was determined to have a symmetrical probability density function. Thus the median filter yielded an excellent estimate of the uncorrupted signal with no degradation of edges or other details. The median filter is compared to a triangular window function.

This publication has 0 references indexed in Scilit: