Color, as a core element in artistic images, not only contributes to visual aesthetics but also plays a crucial role in expressing emotions, constructing thematic narratives, and conveying semantic information. While existing deep learning approaches for art style recognition primarily focus on structural and textural features, they often overlook the perceptual and psychological dimensions of color. To address this gap, this paper proposes a color perception–aware deep learning framework for artistic style classification. Our method integrates psychologically motivated color preprocessing—including saturation enhancement and adaptive transformation in HSV and Lab color spaces—into a standard ResNet-50 backbone, enabling the model to better capture stylistically relevant chromatic cues. Evaluated on the WikiArt dataset (27 styles), our approach achieves a test accuracy of 90.2% ± 0.3, outperforming current state-of-the-art methods without resorting to larger architectures or multi-modal inputs. The results demonstrate that explicitly modeling human color perception can yield significant performance gains in art style recognition using only image data. Beyond classification, we discuss the broader implications of precise color feature extraction for applications such as digital restoration, artwork authentication, and creative content recommendation. Finally, we outline future directions in color-aware AI for art, emphasizing its potential in emotional analysis, generative art, and interdisciplinary research at the intersection of computer vision and art history.
Huichao Zhang (Thu,) studied this question.