Adaptive Reconstruction Levels for Video Quantization

The patent applies local quality by using different quantization step sizes for different sub-parts of the data space. Instead of a uniform quantization step size across all data, the method partitions the data space into multiple sub-parts, each with its own optimized quantization step size. This allows the quantization process to adapt to local characteristics of the data distribution, improving rate-distortion performance while maintaining reasonable device complexity through the structured approach to determining these step sizes.

The patent applies preliminary action by pre-determining a set of candidate quantization step sizes before the actual quantization process. These candidate step sizes are established in advance based on statistical analysis or optimization criteria, so that during encoding, the system only needs to select from this pre-computed set rather than performing complex real-time optimization. This reduces coding time while still achieving improved rate-distortion performance through the use of adaptive step sizes selected from the pre-determined candidates.

The patent applies segmentation by dividing the data space into multiple sub-parts, each characterized by its own data distribution statistics. For each sub-part, a specific quantization step size is determined based on its local characteristics. This segmentation approach allows the system to achieve high reconstruction accuracy by adapting to local data patterns while managing device complexity through the modular structure of processing different sub-parts separately with predetermined step sizes.