Abstract
Detecting spatial targets in complex backgrounds is a fundamental visual task. Detectability decreases with increasing retinal eccentricity. Detectability is also affected by the properties of the background surrounding and under the target, an effect known as masking. Previous research has used additive targets, where the target intensities are added to the background. Previous research shows that background contrast explains much of the variance in foveal and peripheral detectability. Furthermore, the masking of additive targets operates in a lawful manner: Target threshold power increases linearly with background contrast power. In natural viewing, however, targets occlude the background, replacing background features with target features. It is currently unknown whether the laws that apply to additive targets extend to occluding targets. We began our investigation with an experiment using two occluding targets (a uniform disc and vertical edge) embedded in 1/f noise at varying eccentricities. The uniform disc had a luminance of 25 cd/m2. The vertical edge target had the same mean luminance and a contrast of 17% RMS. The average luminance of the background was fixed at 25 cd/m2. An eccentricity psychometric function was measured for five background contrasts (3%, 10%, 17%, 31%, 45% RMS). The eccentricity threshold is the eccentricity where detection accuracy falls to 75%. A lower threshold indicates that a target must be presented closer to the fovea to be detected. We find that eccentricity thresholds decrease with background contrast for the vertical edge target, but increase for the uniform disc target. This is very different from additive targets, where detectability always decreases with background contrast. While this difference makes intuitive sense, it highlights the need for measuring and modeling detection of occluding targets. We are currently extending our measurements to natural backgrounds and are developing a model of detection to account for both additive and occluding targets.
Meeting abstract presented at VSS 2016