Abstract

Previous work assessed the efficiency of temporal integration by human observers for detection of static signals embedded in spatiotemporal (dynamic) noise. In a new paradigm, the signal is spatially displaced form frame to frame. For an ideal observer that integrates the responses of matched filters at different signal locations through time, the performance (d’) remains the same as if the signal were static, improving as the square root of the number of frames in the sequence. We evaluated motion compensated integration by human observers using a 4 AFC detection of a moving (.37 c/frame) Gabor patch (5.5 c/degree) embedded in spatiotemporal Gaussian noise. Observers fixated during the experiment. Performance was measured for increasing number of frames in the sequence (1 to 40). The sequences were displayed at 20 frames/sec; however, each sequence was repeatedly displayed for a total display time of 4 sec. For comparison the experiment was repeated with a non-moving signal embedded in dynamic noise and a signal embedded in static/spatial noise (single frame). Results: For equivalent signal to noise ratios for the ideal observer, the corresponding statistical efficiencies (d’²_human/d’²_ideal) for human observers were approximately 30% for the static noise, 10% for the non-moving signal in dynamic noise and 7% for the moving signal in dynamic noise. The critical time for the motion compensated integration is in the same range as the spatially local temporal integration; 700-1000 msec.

© 1993 Optical Society of America