Exemplar-Based Image and Video Stylization Using Fully Convolutional Semantic Features.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Digital Image Processing

Background:

• Current photo enhancement software lacks semantic understanding, limiting artistic style range.
• Manual style creation is time-consuming and lacks flexibility.
• Artistic stylization requires distinct adjustments for different image regions.

Purpose of the Study:

• To propose a novel deep learning architecture for exemplar-based image and video stylization.
• To enable semantic-aware local enhancement styles.
• To achieve efficient stylization with a single forward pass.

Main Methods:

• Developed a deep learning architecture using fully convolutional networks for feature extraction.
• Integrated fully connected neural layers for adjustment prediction.
• Extended image stylization to videos using temporal superpixels for style transfer.

Main Results:

• The proposed deep learning architecture efficiently performs image stylization in a single forward pass.
• The method successfully transfers artistic styles from image exemplars to videos.
• Experiments demonstrate the effectiveness of the architecture on diverse datasets and video clips.

Conclusions:

• The novel deep learning approach offers a powerful and efficient solution for image and video stylization.
• Semantic-aware local adjustments significantly expand the possibilities for artistic visual styles.
• The method provides a significant advancement over traditional, global color transform techniques.