AI Images vs. Real Photographs: Investigating Visual Recognition and Perception

Abstract

Recently, the photorealism of generated images has improved noticeably due to the development of AI algorithms. These are high-resolution images of human faces and bodies, cats and dogs, vehicles, and other categories of objects that the untrained eye cannot distinguish from authentic photographs. The study assessed how people perceive 12 pictures generated by AI vs. 12 real photographs. Six main categories of stimuli were selected: architecture, art, faces, cars, landscapes, and pets. The visual perception of selected images was studied by means of eye tracking and gaze patterns as well as time characteristics, compared with consideration to the respondent groups’ gender and knowledge of AI graphics. After the experiment, the study participants analysed the pictures again in order to describe the reasons for their choice. The results show that AI images of pets and real photographs of architecture were the easiest to identify. The largest differences in visual perception are between men and women as well as between those experienced in digital graphics (including AI images) and the rest. Based on the analysis, several recommendations are suggested for AI developers and end-users.