Classification Bayesian models of orientation identification with the known reference

Abstract

Humans may integrate the information of cues into target identification. We investigate how better the human brain identifies the orientation in the presence of the known reference. We first design a psychophysical experiment of orientation identification. The subjects estimate the orientation of a line which is intersected by a known oriented reference line. Four subjects performed the identification task. Estimates of the orientations exhibit the systematic increasing biases with the angle between the target line and the reference line increasing, and then the estimation precision of tilt orientations is obviously improved. We expound the identification process by Bayesian inference theory. We assume that the subjects first classify the stimuli and subsequently identify them. Then we put forward two classification Bayesian identification models: Directly Identifying Classification Bayesian Model (DCB) and Indirectly Identifying Classification Bayesian Model (ICB), in which the Equal-precision and Variable-precision encoding are considered. We compare our models' predictions to the experimental data. The results show that the variable-precision indirectly identifying classification Bayesian model fit better to the performance.