This paper describes early progress in novel, interdisciplinary work that applies concepts and methods from Computer Vision to the development of a visual model of restorative scenes. Such a model has the potential to both enhance Attention Restoration Theory (ART) and to find numerous practical applications in the design and synthesis of living spaces and visual artefacts. To explore the feasibility of a visual model of restorative scenes, a comparison between known human gaze patterns and an exclusively bottom-up computational model of attention was performed. Similarities were found, providing evidence for a key claim of ART. Visual Models were then developed with 3 different motivations: i) biological plausibility, ii) a concern for model interpretability, and iii) a hypothesis that more abstract scene properties such as complexity and information content are responsible for fascination and restoration. Image datasets rated by humans for restorative potential were developed and used to construct and test these models and encouraging results were found. This work is the first to combine Computer Vision and Environmental Psychology and it is hoped that further collaborations are inspired.
