Human and Ideal Observers for Detecting Image Curves

摘要: This paper compares the ability of human observers to detect target image curves with that an ideal observer. The are sampled from a generative model which specifies (probabilistically) geometry and local intensity properties curve. observer performs Bayesian inference on using MAP estimation. Varying probability for curve enables us investigate whether performance is best obey specific shape statistics, in particular those observed natural shapes. Experiments performed data both rectangular hexagonal lattices. Our results show observers' approaches are, general, closest conditions where tends be straight or similar statistics curves. suggests bias towards statistics.