Evaluating TNO Human Target Detection Experimental Results Agreement with Various Image Metrics

Abstract : An evaluation of the agreement between experimental results of human target detection performance, as obtained by TNO - Human Factors Research Institute, and various image metrics is addressed in this paper. Image metrics, such as local target from background distinctness metrics (DOYLE and TARGET, a global image complexity metric (POE) and a textural global I local co-occurrence matrix metric (ICOM), are presented and applied to the TNO image database. Good agreement, denoted by relatively high correlation levels, is found between the experimental results (search rates and probabilities of detection) and both DOYLE and TARGET local image metrics values. On the other hand, a relatively low correlation level is obtained between the experimental results and the POE global image metric values. Correlation values obtained using the global I local ICOM metric are between these extremes, as expected. These results emphasize the dominance of the target to background distinctness perceptual cue and the appropriateness of the local metrics to this kind of imagery. Furthermore, they can be used to formulate empirical classification rules that can be used to evaluate and predict human detection performance in similar cases.