Video Super-Resolution via Segmented Integral and Non-Integral Scaling

The patent segments the super-resolution process into two distinct stages: first performing super-resolution processing by integral multiple times to obtain an intermediate high-resolution video image, then performing super-resolution processing by non-integral multiple times on the intermediate image to obtain the final target high-resolution video image. This segmentation allows coordinate mapping to be performed accurately in the first stage with integral multiples, while the second stage handles the remaining non-integral scaling with reduced error propagation.

The patent merges two different super-resolution processing methods: super-resolution processing by integral multiple times (which provides accurate coordinate mapping and reduces blurring) and super-resolution processing by non-integral multiple times (which provides processing simplicity). By combining these methods in sequence, the patent achieves both good image quality and processing efficiency, avoiding the blurring and jagged edges that occur when using interpolation alone for large magnifications.

The patent introduces dynamic adaptability by detecting whether the scaling factor is an integral multiple or a non-integral multiple, and then dynamically selecting the appropriate processing path. When the scaling factor is an integral multiple, the system performs super-resolution processing by integral multiple times with accurate coordinate mapping. When the scaling factor is a non-integral multiple, the system performs super-resolution processing by non-integral multiple times. This dynamic adaptation allows the system to maintain high image quality while being versatile across different scaling scenarios.