“…Despite this, the representation of shape in the PLOS COMPUTATIONAL BIOLOGY human visual system remains elusive, and the basis for shape similarity judgments remains unclear. In part, this is due to the numerous potential shape descriptors proposed in the past, including simple metrics, like solidity [36], and contour curvature [39], and more complex metrics like shape context [38], part-based ones [1,85], Fourier descriptors [41,100,101], radial frequency components [82,102], shape skeletons [40,46,98,99,[103][104][105][106][107], linearity [108] convexity [109][110][111][112], triangularity [113], rectilinearity [114], information content [115,116] and models based on generalized cylinders for describing 3D animal-like objects [117]. While it is widely believed that human shape representations are multidimensional, to date there has been no comprehensive attempt to implement this idea in a concrete image-computable model.…”