Perceived Restorativeness Across Image-Derived Visual Space Types in Urban Parks: A Scene-Based Multilevel Evaluation

Urban parks provide important settings for restorative experiences, yet how perceived restorativeness varies across visually distinct park scenes remains insufficiently understood. This study examined three urban parks in Fuzhou, China. Mask2Former was applied to 368 panoramic images to derive five visual composition indicators, and hierarchical clustering identified four visual space types: High-Vegetation, Waterfront-Open, Built-Interface, and Open-Pathway. Twenty-four selected scenes were evaluated by 300 participants, yielding 2400 participant–scene observations. Mixed-design repeated-measures ANOVA, cross-classified linear mixed-effects models, and relative contribution analysis were conducted. The High-Vegetation type received the highest ratings for perceived restorativeness and space use intention, whereas the Open-Pathway type received the lowest. Small but significant visual space type × questionnaire version interactions indicated some sensitivity to specific scene exemplars. Perceived restorativeness was associated with multiple perceptual dimensions, with Refuge, Culture, Social, and Nature showing relatively prominent contributions in the present sample. Space showed no statistically significant adjusted association in the primary mixed-effects model, whereas Prospect remained positively associated with perceived restorativeness. The study integrates image-derived visual composition, multidimensional environmental perception, and perceived restorativeness within a scene-based analytical framework, providing quantitative evidence for differentiated urban park design considerations.