The efficient organization of the metadata of artwork images necessitates an accurate image classification technique that can identify painting style. Thus, we propose a handcrafted algorithm to extract image features that evaluate brushstrokes and color usage in artwork. The proposed features are based on disparities in color hue between adjacent superpixels, which aim to represent a single brushstroke. Our experiment, which employed 22,314 images of Cubism, Romanticism, and Impressionism art, demonstrates that the proposed features significantly improve the macro-F1 score, which is the overall evaluation metric, by at most 2.96%, compared with the combination of conventional handcrafted features. In addition, both macro-precision and recall are improved by a maximum of 3.03% and 3.02%, respectively. Furthermore, the efficacy of the proposed features is demonstrated both by the increase in the F1-scores for all painting styles and the considerable improvement exhibited in the mean and median mutual information (at least 51.29%). Further comparisons and discussions are provided to verify the effectiveness of the proposed features.
KINOSHITA et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: