Block-level super-resolution video coding

The patent divides the high-resolution video frame into multiple lower-resolution blocks through downsampling, processes each block independently using conventional intra-prediction, and then reconstructs the high-resolution frame. This segmentation allows simple prediction methods to achieve high compression efficiency on complex content by breaking down the difficult task into manageable smaller units.

The patent transforms the problem from high-resolution direct encoding to low-resolution encoding by changing the resolution dimension. By downsampling the high-resolution frame to lower-resolution blocks, the system operates in a different dimensional space where conventional prediction methods are more effective, then maps back to the original high-resolution space through upsampling and residual coding.

The patent segments the high-resolution video block into multiple lower-resolution blocks, processes each with reduced detail, and combines the results. This segmentation reduces the amount of data that needs to be transmitted while maintaining prediction accuracy through the super-resolution reconstruction process that synthesizes high-resolution predictions from multiple low-resolution components.

The patent discards some high-frequency detail information during downsampling to low-resolution blocks, processes these simplified representations, then recovers the high-resolution prediction through super-resolution techniques. The residual coding captures and transmits only the essential high-frequency information that was discarded, achieving compact bitstream representation.