Super-resolution fusion of complementary panoramic images based on cross-selection kernel regression interpolation

Area of Science:

• Computer Vision
• Image Processing
• Optical Engineering

Background:

• Omnidirectional imaging often suffers from low and nonuniform resolution.
• Complementary catadioptric imaging techniques aim to address these limitations.
• Generating high-resolution panoramic images from omnidirectional views remains a challenge.

Purpose of the Study:

• To develop a method for generating high-resolution panoramic images from omnidirectional views.
• To fuse complementary inner and outer images without interference.
• To enhance the quality and resolution of panoramic images.

Main Methods:

• A cross-selection kernel regression method is proposed for image fusion.
• Horizontal gradients are estimated using the outer image's scattered pixels.
• Vertical gradients are estimated using the inner image.
• Kernel regression adaptively steers based on local gradients and pixel selection.

Main Results:

• The proposed method successfully fuses complementary inner and outer images.
• It effectively estimates horizontal and vertical gradients for panoramic image generation.
• Adaptive kernel steering and pixel selection improve interpolation accuracy.

Conclusions:

• The cross-selection kernel regression method significantly outperforms existing techniques.
• It achieves superior visual quality and objective evaluation metrics.
• This approach offers a robust solution for high-resolution panoramic image generation.