Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Using Artificial Neural Networks to Understand Fluency in the Perception of Paintings

Yi Lin¹, Johan Wagemans¹, Hans Op de Beeck¹; ¹KU Leuven

Presenter: Yi Lin

Fluency—the ease with which an image is processed—is closely linked to aesthetic appraisal. However, existing objective measures of fluency fail to accurately capture subjective ratings. Recently, deep convolutional neural networks (DCNNs) have been proposed as a tool for measuring integration, a concept linked with fluency, using natural scenes as stimuli. Yet, the direct link between fluency and integration remains untested, and it is unclear whether these findings generalize to art perception. In this work, we investigate (1) whether the integration measure via DCNNs effectively captures fluency, potentially outperforming existing methods in the context of art perception, and (2) if DCNNs provide a superior measure of fluency, what specific mechanisms they reveal. Our findings indicate that the DCNN-based integration measure captures subjective fluency well and significantly outperforms other objective fluency measurements. Additionally, we observed that the peak correlation between DCNN-derived integration and various visual characteristics—intended to quantify different aspects of fluency—occurs at different DCNN layers. This suggests that fluency may be a multi-level process, integrating distinct visual characteristics at various processing stages. In summary, a DCNN-based measure of integration provides valuable insights into the concept of fluency.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF