Creating Perceptually Uniform Circular Shape Spaces: An Automated Method for Asymmetric Naturalistic Shapes

Shape processing is crucial for higher-level visual functions such as object recognition. Yet, isolating high-level vision from low-level shape influences remains challenging. Typically, studies rely on simple, symmetric stimuli and labor-intensive psychophysical measurements to control for perceptual shape similarity. However, natural objects are often asymmetric, and researchers currently lack efficient methods or validated stimulus sets for these complex shapes. Here, we present an image-based method that autonomously generates perceptually uniform shape sets composed of novel, asymmetric, animal-like shapes. We focus on circular shape spaces, where the angular distance between two shapes corresponds to their perceived similarity. Our method employs a search for circular arrangements of shapes that maximize perceptual uniformity according to ShapeComp, an image-computable shape similarity metric based on 100+ descriptors that is highly predictive of human shape similarity for naturalistic shapes. Through multi-arrangement experiments, we validated that our method produces circular shape sets that align with human similarity judgments. By comparison, uniform and circular shape sets according to alternative shape spaces (e.g., Generative Adversarial Networks or Radial Frequency patterns), but not according to ShapeComp, were not necessarily perceived as uniform and circular. We provide five newly validated circular shape sets derived from intricate naturalistic shapes, along with MATLAB code facilitating the generation of an unlimited number of sets that can be tailored to experimental needs beyond circular configurations. By offering efficiency and adaptability, our method allows to systematically explore higher-level influences on object perception while minimizing the time and resources required to construct controlled, complex shape stimuli.