CG2Real: Improving the Realism of Computer Generated Images Using a Large Collection of Photographs.

Area of Science:

• Computer Graphics
• Image Processing
• Artificial Intelligence

Background:

• Computer-generated (CG) images strive for realism but often remain distinguishable from real photographs.
• Manual modeling for realistic CG imagery is time-consuming and costly.

Purpose of the Study:

• To develop a data-driven approach for enhancing the realism of CG images.
• To create a fast and easily integrated system for improving digital content.

Main Methods:

• A novel approach using a large dataset of photographs from online repositories.
• Retrieval of real images with similar global structure to a given CG image.
• Identification of corresponding regions using mean-shift cosegmentation for automatic color, tone, and texture transfer.

Main Results:

• The developed system successfully transfers realistic elements from real images to CG images.
• User studies indicate that the resulting hybrid images are perceived as more realistic than the original CG images.
• The system operates using only image processing, eliminating the need for 3D models.

Conclusions:

• The proposed data-driven method offers an efficient and effective way to increase the realism of CG imagery.
• This technique integrates seamlessly into existing digital content creation workflows.
• The findings suggest a promising direction for future research in realistic image synthesis.