Image Stylization Using Structural and Texture Feature Fusion

The patent introduces a pre-trained encoder as an intermediary component that automatically extracts structural features from images. This encoder serves as a bridge between raw image data and the style transfer process, enabling comprehensive content processing without manual intervention. The encoder processes both target objects and backgrounds uniformly, ensuring complete stylization while maintaining processing efficiency through automated feature extraction.

The system employs a pre-trained encoder that performs self-service by automatically extracting structural features from images without requiring manual drawing of stylized samples. The encoder is pre-trained on general image data and can be directly applied to extract features from any input image, eliminating the need for time-consuming manual sample creation while maintaining high processing accuracy through its learned feature representation capabilities.

The patent implements a universal style transfer approach where the pre-trained encoder and subsequent processing steps apply uniformly to both target objects and backgrounds. The encoder extracts structural features from the entire image without discrimination, and the style transfer mechanism processes all regions consistently. This multi-functional approach ensures that both foreground objects and background elements receive comprehensive stylization, achieving complete image transformation.