Broad brand spatial filters in the human visual system

Binocular summation in orientation discrimination depends on stimulus contrast and duration

Binocular summation, an improvement in visual performance with binocular viewing compared to monocular viewing, has been studied extensively in detection tasks. Monocular detection thresholds for stationary stimuli are typically about 40% higher than binocular thresholds. Binocular summation in discrimination tasks, however, is often lower and less consistent. A possible explanation for this difference is that saturation of responses limits the extent of binocular summation in discrimination tasks. To investigate this possibility, we used an orientation discrimination task and varied stimulus contrast and exposure duration. Monocular and binocular orientation discrimination thresholds were obtained using one-dimensional difference-of-Gaussian stimuli. For briefly exposed stimuli, binocular summation is greatest at low contrasts (e.g. 66% at 8% contrast) and is reduced systematically at higher contrasts so that monocular and binocular thresholds are approximately equal at contrasts above 15%. Binocular summation for low-contrast stimuli is greatest at a brief exposure duration (50 msec), is reduced at longer durations, and is not significant at durations of 100 msec or longer. Thus, binocular summation in orientation discrimination is greatest for relatively low-energy stimuli. These results are consistent with models of binocular energy summation and the hypothesis that saturation of responses after binocular combination can limit binocular summation in discrimination tasks.

Entropy-based texture analysis in the spatial frequency domain

An approach which uses regional entropy measures in the spatial frequency domain for texture discrimination is presented. The measures provide texture discriminating information independent of that contained in the usual summed energy within based frequency domain features. Performance of the entropy features as measured by a between-to-within-class scatter criterion is comparable to that of traditional frequency domain features and gray level co-occurrence contrast features. A method of frequency scaling is introduced to enable the comparison of texture samples of different subimage size. The resulting regional entropy measures are subimage size-invariant subject to certain constraints which arise from properties of the discrete Fourier transform.