Video Decoding Quantization Matrix Selection to Reduce Bit Overhead

The patent segments the quantization process by dividing transform blocks into different frequency regions (e.g., high-frequency and low-frequency regions). Different quantization matrices are applied to different regions, allowing frequency-dependent scaling while managing complexity through selective application rather than universal processing of all coefficients with multiple matrices.

The patent applies different quantization characteristics to different frequency regions within transform blocks. High-frequency coefficients use one quantization matrix while low-frequency coefficients use another, allowing local optimization of compression quality in different frequency domains without requiring the decoder to process all possible matrices uniformly across the entire block.

The patent extracts and transmits only the essential quantization matrix information needed for specific frequency regions rather than encoding and transmitting complete sets of multiple full matrices. By taking out only the necessary components (e.g., differential values or selected matrix elements), the patent reduces bit overhead while preserving the ability to apply multiple quantization matrices where needed.

The patent applies partial action by using multiple quantization matrices only for specific frequency regions rather than applying them universally to all transform coefficients. This selective application provides encoding flexibility for regions where it matters most while avoiding the bit overhead associated with transmitting and processing complete multiple-matrix sets for entire blocks.