Recognition and Transformation of Style Features in Modern Architectural Images

This study presents a novel approach to style recognition and transformation in modern architectural images. In the era of digital art, the demand for efficient creation of high-quality digital images has led to the emergence of style transfer technology. However, existing methods face challenges when processing images with complex semantic information, such as architectural scenes. This research proposes a deep learning-based solution to accurately identify and extract semantic regions within images, enabling more refined style transfer. The methodology involves efficient image feature extraction using a simplified ResNet50 network with a feature extraction module, followed by a transformation process that maintains style consistency across regions while accounting for stylistic variations. A comprehensive loss function, including content, style, and regularization terms, guides the style transfer process. Experimental results demonstrate the effectiveness of the proposed approach in preserving image semantics and improving style transfer quality.