Which shape features determine detectability of camouflaged radial frequency patterns?

Poster Presentation: Tuesday, May 21, 2024, 2:45 – 6:45 pm, Pavilion
Session: Object Recognition: Basic features

Wei Hau Lew1,3 (), Samuel P. Smithers2, Peter J. Bex2, Daniel R. Coates1,4; 1University of Houston College of Optometry, Houston TX, 2Department of Psychology, Northeastern University, Boston MA, 3Schepens Eye Research Institute of Mass Eye and Ear, Department of Ophthalmology Harvard Medical School, 4University of Washington, Department of Ophthalmology, Seattle WA

When the texture or pattern of stationary targets is closely matched to the background, the detection of these camouflaged targets relies on identifying specific features such as the edge of its outline. Previously, we reported the results of behavioral experiments that tested the detection time of targets defined by radial frequency (RF= 3, 4, 5, 8, 10, 14, and 20) and amplitude (0.1, 0.25, and 0.5) with a pink noise (1/f) texture against a pink noise background. At low amplitude, mid RF targets (i.e. 8 and 10) were among the hardest to detect. For RF≥ 8, the time taken to detect the targets decreased (targets became easier to detect) as amplitude increased. Since it is still poorly understood what aspects of a target shape influence target detectability, here the local and global features of each shape were analyzed. We calculated the global features of perimeter, area, and aspect ratio (width/height of each shape “arm”). The osculating circle method was used to compute the local curvature, its Sum-Squared Difference (SSD), and Mean of Absolute Difference (MAD). We found that target detection time as a function of the perimeter had an inverted U-shape. The area did not systematically affect detection time. The aspect ratio also had an inverted U-shape, whereby targets were easier to detect when their arm’s aspect ratio was <2. For target curvature, there was a positive correlation between SSD of the curvature and target detectability (Pearson’s r= 0.844, p <0.001), but no correlation for MAD. While the local SSD was directly correlated and most predictive, global features (perimeter and aspect ratio) can also influence detectability.