SC-AOF: A Sliding Camera and Asymmetric Optical-Flow-Based Blending Method for Image Stitching

Area of Science:

• Computer Vision and Image Processing
• Computational Photography

Background:

• Image stitching commonly faces challenges with parallax and perspective deformation.
• Existing methods struggle to maintain structural integrity and achieve seamless transitions.

Purpose of the Study:

• To develop an advanced image stitching method addressing parallax and perspective deformation.
• To improve structure preservation and natural visual transitions in stitched panoramas.

Main Methods:

• A sliding camera approach with virtual camera interpolation to minimize perspective deformation.
• Bidirectional asymmetric optical flow calculation using gradient descent for parallax reduction.
• Softmax function and registration error for adaptive blending area adjustment to eliminate ghosting.

Main Results:

• The proposed method effectively reduces perspective deformation and parallax.
• Achieved more natural image transitions and superior local alignment compared to state-of-the-art methods.
• Demonstrated higher structural similarity index and robust performance across various scenarios.

Conclusions:

• The novel image stitching technique successfully overcomes limitations of previous methods.
• It offers a robust and visually superior solution for creating high-quality panoramic images.
• The method provides a significant advancement in parallax processing and structure preservation.