Human and monkey detection performance in natural images compared with V1 population responses

Abstract

Detection is a fundamental task that is critical to visual behavior. The central aim of this study was to measure and model behavioral and neurophysiological performance for detecting targets under naturalistic conditions. I first measured behavioral detection performance macaques and compared it to humans. Detection thresholds were measured on uniform backgrounds and for several contrasts of natural image backgrounds. I find that (i) threshold contrast power is a linear function of background contrast power for both humans and macaques, and (ii) the relative threshold functions for humans and macaques are in good agreement, although (iii) the macaques are less sensitive overall. Subsequently, I investigated the quantitative relationship between V1 population responses and detection performance. I used voltage-sensitive dye imaging (VSDI) to measure the neural population activity in V1 for the same stimuli, while the monkeys held fixation. The spatial scale of VSDI measurements was sufficient to resolve retinotopic responses and orientation columns over the whole region activated by the target. Separate read-out strategies were used for retinotopic and columnar responses. Across multiple contrast levels of natural image backgrounds, I compared both scales of population responses between target-present and target-absent conditions to derive the signal-to-noise ratio (d’), which specifies neurometric functions. Based on this simple approach, the results show that in comparison to behavioral performances, retinotopic performances degraded at a relatively higher rate with increasing contrast masking. On the other hand, columnar performances were relatively less susceptible to contrast masking in natural image backgrounds.