Evaluating human detection performance of targets and false alarms, using a statistical texture image metric

A statistical texture image metric, which is based on the Markov cooccurrence matrix and named ICOM, is introduced and evaluated in the context of the correlation between (1) human quantitative detection performance of targets and false alarms and (2) qualitative image judgments of both natural and enhanced infrared images. Correlations between the ICOM metric and experimental detection results are compared with those obtained with the probability of edge (POE) global clutter metric, the local contrast metric (DOYLE), and the combined signal-to-clutter (SCR) metric. In contrast with those metrics, which fit well either the qualitative or the quantitative results, the ICOM textural metric was found in good agreement with both the qualitative and quantitative experiments. Moreover, the ICOM metric was found appropriate for automatic extraction of potential false targets in both natural and the enhanced images. This property is used to analyze human false-detection-decision behavior, and to suggest a modification to the known constant false-alarm rate (CFAR) model. The modified model considers the total number of detection decisions (true and false) made by the human observer as the adaptive parameter. The model was tested and confirmed both with natural and enhanced images. The ICOM properties and the results obtained using them emphasize the robustness and the adequacy of this metric for infrared imagery evaluation.