This technology is an exemplar-based inpainting device and method that projects each location of a spherical panoramic video onto a spherical surface based on latitude and longitude, standardizes them into uniform square patches regardless of latitude, and then identifies and copies reference patches most similar to the priority pixels at the boundary of the damaged area.
Spherical panoramic videos are 2:1 rectangular projections of a spherical coordinate system, which causes subjects to stretch horizontally at higher latitudes, leading to significant errors when using standard planar image restoration techniques. Existing exemplar-based inpainting was limited to areas near the equator with minimal distortion, while pixel-based restoration methods struggled to reconstruct large damaged areas or repetitive patterns.
This technology uses a coordinate transformation formula to project onto a virtual sphere to obtain patches, then iteratively determines priorities, searches for similar patches, and copies pixels until the target area is fully restored, ensuring natural results that account for latitude-based distortion. It can be applied to 360-degree camera-based AR content and virtual tour video editing to seamlessly remove markers or unwanted objects from omnidirectional footage.
N/A