Abstract: A system and method is disclosed for detecting whether a photograph is of a flat surface, e.g. a “photograph of a photograph”, or of a real three dimensional scene. The method includes using a digital camera to take a photograph, and within a predetermined time period either before or after taking the photograph, recording focal length information from a plurality of focus areas using the autofocus routine of the digital camera. The recorded focal length information forms the input to a machine learning classifier which has been trained to classify scenes as “flat” or “3D”.