Selective Video Frame Filtering for Decoding Complexity Reduction

The patent applies different filtering operations to different regions of the video frame based on their characteristics. Specifically, it identifies important regions (such as regions containing important objects or high-activity areas) and applies deep learning-based loop filtering only to these regions, while using conventional filtering or no filtering for other regions. This selective approach maintains video quality in critical areas while significantly reducing the overall computational burden and decoding complexity.

The patent segments the video frame into multiple regions based on importance, activity level, or other criteria. By dividing the frame into different zones, the system can apply appropriate filtering strategies to each segment independently, rather than uniformly processing the entire frame with computationally intensive deep learning filters.

The patent implements partial filtering by selectively applying deep learning-based loop filtering only to important regions of the video frame rather than all blocks. The system identifies regions that benefit most from filtering and concentrates computational resources there, while using simpler or no filtering for less critical areas, thereby avoiding waste of computational resources on blocks with minimal gains.

The patent changes the filtering parameters (such as filter strength, filter type, or whether to apply filtering at all) based on the characteristics of different regions. By dynamically adjusting these parameters according to regional importance or content characteristics, the system optimizes the balance between filtering effectiveness and computational resource consumption.