DST7-Based Multiple Transform Selection for Video Decoding Efficiency

The patent replaces the traditional mechanical computation of DST7 with a substitution using Discrete Fourier Transform (DFT) operations. By expressing DST7 in terms of DFT, the implementation leverages existing efficient DFT algorithms and libraries, reducing computational complexity while maintaining transform accuracy. This substitution allows using mature, optimized DFT implementations rather than re-inventing DST7 computation.

The patent changes the parameter representation by introducing scaling factors that convert DFT results into DST7 output. By using scaling parameters (such as scaling factors of 1/8, 1/16, or 1/32 depending on transform size), the system achieves exact DST7 results through parameter transformation rather than direct computation, reducing operational complexity while preserving precision.

The patent implements dynamics by allowing the transform type to be dynamically selected based on coding conditions. The encoder/decoder can adaptively choose between DCT2 and DST7 transforms depending on the specific coding scenario, block characteristics, or configuration parameters. This dynamic selection provides coding flexibility without requiring support for all possible transforms simultaneously, managing complexity through conditional application.

The patent achieves universality by designing a unified transform framework where both DCT2 and DST7 can be handled through a common processing architecture. The encoder/decoder structure is designed to accommodate multiple transform types using shared components, reducing overall complexity while maintaining the ability to perform different transforms as needed.

The patent applies segmentation by dividing the transform processing into manageable units. Instead of implementing a single large transform for the entire video frame, the system processes smaller transform blocks (e.g., 4x4, 8x8, 16x16) independently. This segmentation reduces memory requirements for each individual transform operation while maintaining the ability to handle high-resolution video through hierarchical processing.

The patent implements nesting by organizing transform operations in a hierarchical structure where smaller transforms are nested within larger ones. The transform processing can be recursively applied at different resolution levels, with smaller transform units nested within larger blocks. This nested approach allows efficient memory management and processing capability scaling for high-resolution video content.