Convolutional Neural Network-based Approach for Painting Style Transfer with Enhanced Content Preservation
Zhifen Zhou · 2025
Painting style transfer is a newly developing field in computer vision that makes it possible to transfer digital images according to the artistic features of different painting styles. Conventional style transfer techniques, though visually appealing, sometimes have problems, including loss of content details, poor generalization across heterogeneous styles, and high computational costs. These disadvantages lead to very serious challenges in obtaining both artistic precision and real-world efficiency in applications. To address these difficulties, this paper suggests a new painting style transfer method using Convolutional Neural Networks (CNNs). The suggested model is tailored to efficiently capture and combine content and style features while maintaining structural coherence and boosting stylistic diversity. Through the use of deep hierarchical convolutional layers, the model attains a harmonious equilibrium between content retention and artistic stylization. A large dataset with pairs of contents and styles of different images was utilized to train and test the model, aiming to maintain robustness across various painting styles and levels of image complexity. Results of experiments prove that the novel CNN-based approach outperforms conventional methods by achieving accuracy of 0.982, precision of 0.981, recall of 0.98, specificity of 0.98 and F1-score of 0.98 in terms of visual quality, structural coherence, as well as computational time by considerable margins.