Enhancing Artistic Portrait Colors Using Deep Learning and Style Transfer: An Information Literacy Approach

Artistic portrait enhancement is an essential technique for the digital preservation of Chinese cultural heritage, since the paintings and drawings are often damaged by color fading, pigment degradation and aging, and the current deep learning technique has not been able to simultaneously preserve the facial identity, structural consistency, texture realism, and artistic style. This paper presents a deep learning-based method to improve the visual quality of Chinese fine-art portraits while preserving the original artistic properties. The proposed framework involves the Chinese Fine Art/Famous Chinese Paintings Dataset from Kaggle, RetinaFace, Bilateral Filtering for edge preserving noise reduction, Local Binary Patterns (LBP) for extracting texture features, VGG-19 for content–style fusion and refinement using Neural Style Transfer (NST) and Gradient Based Neural Style Transfer Optimization (GB-NSTO), and Adaptive Histogram Based Color Enhancement (AHCE) for luminance and color enhancement. The results of the experimental evaluation prove the effectiveness of the proposed framework, with a Structural Similarity Index (SSIM) of 0.92, a Peak Signal-to-Noise Ratio (PSNR) of 28.5 dB, a Fréchet Inception Distance (FID) of 45.3 and a Colorfulness Index of 62.7, as well as a reduction in inference time compared with baseline approaches. They produce visually balanced outputs suited for digital restoration, maintain artistic fidelity, improve texture realism and preserve facial identity, while maintaining enhanced portraits. In summary, the proposed method is well suited to improve Chinese fine-art portraits, since it achieves a good balance between structural preservation, artistic style transfer and color enhancement, and is a feasible method for digital cultural heritage restoration and a potential method for other cultural heritage image enhancement applications.