Towards a model of expectation-driven perception

Abstract

Human perception is an active process by which meaningful information is gathered from the external environment. Application areas such as human-computer interaction (HCI), or the role of human experts in image analysis, highlight the need to understand how humans, especially experts, use prior information when interpreting what they see. Here, we describe how CHREST, a model of expert perception, is currently being extended to support expectation-driven perception of bitmap-level image data, focusing particularly on its ability to learn semantic interpretations.