Emotional Perception Differences in the Digital Representation of Watercolor Diffusion Traces

Digital image processing has made changes to the way of preserving and presenting traditional watercolor paintings. Diffusion trace effects, such as edge bleeding, pigment deposition, color gradation, and irregular texture caused by watercolor diffusion might alter during the process of digital conversion and affect viewers' emotional response to a certain degree. Twenty watercolor images having distinct diffusion traces were chosen to form four types of visual stimuli: original images, smoothed images, texture enhanced images, and digitally reconstructed images. Ratings for valence and arousal were acquired through SAM. Color variation, local entropy, edge density, texture contrast, and diffusion area were extracted through OpenCV, while deep visual features were obtained through ResNet50. Then, XGBoost and SHAP were performed for emotion prediction and explanation of contribution. It is revealed that the average arousal of smoothed images decreased from 5.31 ± 1.16 to 4.80 ± 0.99, while the arousal of texture enhanced images increased to 5.92 ± 1.12. As for digitally reconstructed images, their arousal was 5.37 ± 1.19. The MAE, RMSE, and R² for fused feature model was 0.60, 0.80, and 0.64. Local entropy and edge density made significant contributions, which implies that diffusion textures and edge structure are highly correlated with emotional arousal in digital watercolor. These conclusions provide a quantitative base for the assessment of digital watercolor, computational art study, and affective image processing.