Video Deblurring via Hidden Variable Fusion

The patent applies local quality by processing different regions of the video frame with different operations based on their blur characteristics. The method segments the frame into multiple regions and applies region-specific deblurring operations, allowing accurate pixel displacement prediction within each region while maintaining overall deblurring quality despite inconsistent blur degrees across the entire frame.

The patent segments the video deblurring process into multiple independent modules: feature extraction, hidden variable prediction, and additive fusion. Each module processes information separately and contributes to the final result, enabling accurate displacement prediction in one module while maintaining overall deblurring quality through the coordinated operation of all modules.

The patent performs preliminary feature extraction and hidden variable acquisition before the main deblurring operation. By pre-processing the video frames to extract intrinsic features and compute forward and backward hidden variable sets, the system reduces the computational burden during the actual deblurring process, improving deblurring effect while managing computational complexity through staged processing.

The patent merges multiple computational operations into a unified additive fusion process. Instead of performing separate deblurring operations for each region or frame, the method combines feature extraction, hidden variable prediction, and result integration into a single coordinated process, reducing overall computational complexity while maintaining high deblurring quality.

The patent implements feedback mechanisms where the system continuously refines its predictions by comparing intrinsic features with hidden variable sets from adjacent frames. The additive fusion process incorporates feedback loops that adjust the deblurring results based on the consistency of information across frames, improving information utilization while filtering out wrong information through iterative verification.

The patent introduces hidden variable sets as intermediary representations that mediate between adjacent frames and the current frame being processed. These hidden variables serve as a buffer that filters and transforms information from adjacent frames, allowing the system to utilize valuable information while preventing wrong information from directly affecting the deblurring result.