Geometrically Consistent Stereoscopic Image Editing Using Patch-Based Synthesis.

Area of Science:

• Computer Vision
• Image Processing
• Computer Graphics

Background:

• Stereoscopic 3D media editing presents challenges in maintaining depth interpretation and scene depth controllability.
• Existing methods often struggle with the inherent correlation between stereo views.

Purpose of the Study:

• To develop a patch-based synthesis framework for advanced stereoscopic image editing.
• To address the core challenges of depth interpretation and scene depth control in 3D media.

Main Methods:

• Introduced a depth-dependent patch-pair similarity measure to differentiate and utilize image content based on depth structures.
• Proposed a joint patch-pair search strategy to effectively manage the correlation between the two stereo views.
• Developed a patch-based optimization framework for stereoscopic image editing.

Main Results:

• Successfully maintained depth interpretation and provided controllability of scene depth in edited stereoscopic images.
• Demonstrated effectiveness across various applications including depth-guided texture synthesis, NPR, paint by depth, content adaptation, and 2D to 3D conversion.
• User studies confirmed the production of stereoscopic images with good stereoscopic and visual quality.

Conclusions:

• The proposed patch-based framework offers robust solutions for a wide range of stereoscopic image editing tasks.
• The novel similarity measure and search strategy significantly improve the handling of depth information and view correlation.
• This work advances the capabilities for editing and creating high-quality stereoscopic 3D content.