Abstract
<p>A salient aspect of our daily visual experiences is the aesthetic impressions. Why do we enjoy some views but not others? Researchers have found visual attributes that we enjoy seeing for colors, shapes, and objects, and even for human faces and bodies; however, aesthetic preferences for holistic visual experiences—views that encompass many different objects and attributes—remain largely undescribed. What principles govern holistic visual aesthetics beyond preferences for certain objects or features? We found computational evidence in support of the prototype principle for visual aesthetics. We measure each view’s prototypicality based on its average similarity to a large reference set of images in feature spaces from pretrained deep convolutional neural networks (AlexNet and VGG-16). For views consisting of inanimate content, the more prototypical the high-level visual representations are, the more aesthetically pleasing the views appear. This prototype effect suggests a functional answer to why we like what we like: We like typical visual experiences that are common, safe, or neutral. Thus, aesthetic preferences from holistic visual experiences are systematic, explainable, and reflect the underlying organization of visual representations of the world. Our method demonstrates a new way to explore both vision and aesthetics using computational models—besides investigating specific visual properties or isolated objects, we can also explore the organization of visual representations and aesthetic experiences holistically, getting us one step closer to understanding how we see, what we like, and why we like what we like in everyday life.</p>