Switched Interpolation Filter Coefficients in Video Bitstream

The patent implements dynamic filter selection by allowing the encoder to switch between multiple interpolation filters (e.g., bilinear, bicubic, Lanczos) based on content characteristics. The decoder receives syntax elements indicating which filter was used, enabling adaptive quality improvement without requiring complex real-time filter design. This resolves the contradiction by making the filter system dynamic rather than static, improving coding quality while maintaining reasonable device complexity through predefined filter options.

The patent changes the parameter of interpolation filter type from fixed to variable. By introducing syntax elements in the bitstream that specify which interpolation filter to use for different blocks or regions, the system adapts filter parameters (filter order, kernel type) to match content characteristics. This allows high-quality reconstruction when complex filters are appropriate while maintaining simplicity for homogeneous regions, thus improving coding quality without proportionally increasing device complexity.

The patent applies adaptive interpolation filters selectively rather than uniformly across the entire video stream. By using content-based decision logic to apply complex filters only in regions where they provide measurable quality improvement (e.g., high-frequency regions, motion boundaries), the system achieves better reconstructed image quality while minimizing the overhead from syntax elements. This partial application strategy resolves the contradiction by avoiding excessive filter adaptation where it is not needed.

The patent segments the video content into different regions or blocks that may use different interpolation filters. By dividing the picture into coding units and independently selecting filters for each unit based on local characteristics, the system improves overall image quality while controlling bitstream overhead. The segmentation allows syntax elements to be sparsely distributed only where filter switching occurs, rather than requiring overhead for every pixel or block.

The patent pre-defines multiple interpolation filter kernels (bilinear, bicubic, Lanczos, etc.) during system initialization or firmware loading, rather than computing them in real-time. These predefined filters are stored in lookup tables or coefficient memory, allowing the encoder to switch between them by selecting different syntax elements. This preliminary preparation provides high adaptability for different content types while keeping runtime device complexity manageable, as the heavy computational work is done in advance.

The patent uses copied or replicated filter coefficient sets for different interpolation types rather than implementing unique computation logic for each filter. By storing predefined coefficient tables for various filter types and simply selecting which table to use, the system achieves versatility in handling different content types while reducing implementation complexity. This copying approach allows multiple filter options to coexist in the device with minimal additional hardware or software overhead.