Mapping image regions for meaning and aesthetics
In a pioneering paper (published in Nature Human Behaviour in 2017) Henderson & Hayes have mapped the semantic richness of local regions in images of natural scenes and demonstrated their role in image exploration by eye movements, compared to pure bottom-up visual saliency. In a first study of this PhD, we will conduct a similar study with three important extensions: we will map image regions of (1) variable size in a nested fashion, (2) for both meaning and aesthetics, and (3) for images of natural scenes as well as for images of art (mainly paintings). We will also compare these maps to saliency maps. In a second study, we will then compare these data to the eye-movement data (fixations, saccades, heatmaps, scan paths) for human observers looking at these images for variable durations and tasks. We will also compare these human eye-movement data to DeepGaze maps. Third, in addition to universal (average) data, we will also investigate individual differences (e.g., as a function of local versus global viewing styles, as a function of expertise for art) and explore how these relate to visual literacy. As a fourth chapter, we will write a theoretical review paper on local and global processing of natural images of scenes and art (both images and real artworks). Finally, based on the previous studies, we will develop (ideas for) a technique to enhance the aesthetic enjoyment of images (and visual literacy) by presenting the right image patches (size and resolution) at the right time, in a particular sequence (corresponding to the ‘ideal’ scan path according to artist and expert data).