Image Processing Quantization Matrix Up-Conversion to Limit Coding Overhead

The patent divides the quantization matrix into multiple sub-blocks corresponding to different frequency regions. Each sub-block can be independently coded or selected, allowing the system to handle large transform units without proportionally increasing the overall coding amount. This segmentation enables selective application of different quantization matrices to different regions.

The patent applies different quantization matrices to different sub-blocks of the transform unit based on their frequency characteristics. High-frequency regions use different matrices compared to low-frequency regions, optimizing compression while maintaining quality. This local differentiation allows efficient handling of large transform units without uniform increase in coding overhead.

The patent segments the quantization matrix selection process into multiple candidate lists, where each list contains matrices suitable for different transform unit sizes or frequency regions. This segmentation allows the decoder to select from predefined groups rather than evaluating all possible matrices, reducing computational complexity while maintaining adaptability.

The patent pre-prepares multiple candidate quantization matrices in advance and organizes them into lists before the actual decoding process. These candidate matrices are arranged according to different transform unit sizes and frequency characteristics, allowing the decoder to quickly select the appropriate matrix without performing complex real-time optimization, thus reducing device complexity.

The patent divides the large quantization matrix into multiple smaller sub-blocks that can be independently processed. This segmentation allows the system to apply appropriate quantization precision to each sub-block based on its frequency characteristics, maintaining overall precision while reducing the total information loss that would occur with a single large matrix applied uniformly.

The patent applies different quantization precision levels to different regions of the transform unit. High-frequency sub-blocks use coarser quantization while low-frequency sub-blocks use finer quantization, optimizing the balance between precision and information loss. This local quality approach ensures that precision is maintained where needed while minimizing overall compression loss.