Abstract
Previous work (Kersten, 1987) has shown that the detection of band-pass noise patterns is well described by an ideal observer, indicating that observers can integrate spatial frequency information efficiently over a six-octave wide band. One interpretation of this result is that observers use a channel with an adjustable bandwidth (Green, 1960) that matches the bandwidth of the signal when detecting band-pass noise. To investigate the notion that observers use adjustable bandwidth channels for spatial frequency, we had observers perform a noise detection task under two conditions: an uncertainty condition where the bandwidth of the noise could vary from trial to trial and a blocked condition where the bandwidth of the signal was held constant during a block. We used horizontal, one-dimensional, band-pass noise patterns that were Gaussian windowed. The center-frequency of the noise was 5 cycles/degree and bandwidth varied from one-half to four octaves. Seven bandwidths were used and a detection threshold measured at each bandwidth for both the blocked and uncertainty conditions. Stimuli were presented for 200ms. At each bandwidth, three 150 trial thresholds were collected. Noise detection r.m.s. contrast thresholds increase with the fourth-root of bandwidth for the ideal observer. For our blocked condition, we again found that human observers' noise detection thresholds increase with the fourth-root of bandwidth (Kersten, 1987). Under conditions of bandwidth uncertainty, we found that detection thresholds continued to increase with the fourth-root of bandwidth. Our results support the notion that when detecting wide-band noise patterns, observers can adjust the band of spatial frequencies they use from trial to trial and select the frequency band efficiently. To explore adjustable channels further, we are investigating the effects of stimulus duration, center-frequency uncertainty and the combination of center-frequency and bandwidth uncertainty on noise detection.