HDR Graphics Compression Endpoint Quantization

The patent segments the HDR pixel data into multiple planes (e.g., R, G, B, A channels) and processes each plane independently through the compression pipeline. This segmentation allows the system to apply quantization and endpoint selection to each plane separately, maintaining precision where needed while achieving compression efficiency. The segmented approach enables parallel processing and efficient memory packing without sacrificing overall bit-depth resolution.

The patent dynamically changes the quantization parameters and endpoint values based on the local characteristics of the pixel block. By adapting the number of bits used for endpoints and weights to the actual content complexity and luminance range, the system maintains high bit-depth resolution in critical areas while allowing greater compression in less sensitive regions. This parameter adaptation resolves the contradiction by making precision allocation flexible rather than fixed.

The patent applies partial quantization rather than full quantization across all pixel data. By using endpoint selection that retains only the most significant bits and applying quantization selectively based on local image characteristics, the system maintains sufficient bit-depth resolution for visual quality while reducing the total number of bits that need to be stored and transmitted. This partial action approach achieves the right balance between precision and bandwidth consumption.

The system dynamically adjusts quantization parameters including the number of bits for endpoints and weights based on content analysis. In regions with smooth gradients or limited luminance range, fewer bits are allocated, reducing memory bandwidth requirements. In regions requiring higher fidelity, more bits are preserved. This adaptive parameter change allows the system to optimize the trade-off between visual quality and bandwidth usage on a per-block basis.

The patent transitions from the traditional single-dimension approach of storing each pixel component with fixed bit-depth to a multi-dimensional representation using planes, endpoints, and weights. This dimensional transformation allows HDR data with variable bit-depth (10-bit, 12-bit, or higher) to be packed into memory structures designed for 8-bit data. The endpoint-weight format effectively creates a new dimensional space where compression efficiency and dynamic range coexist.

The system changes the fundamental parameters of how pixel data is represented and packed by introducing endpoint selection and weight-based compression. Instead of storing full high-bit-depth values for every pixel, the system stores compressed representations using fewer bits for endpoints and weights, achieving memory packing efficiency comparable to 8-bit systems while preserving the ability to represent HDR dynamic range through the mathematical reconstruction of original values.

The patent extracts the essential visual information from high bit-depth HDR data by identifying and storing only the critical endpoints and weights needed to reconstruct the image. Rather than storing all high-bit-depth pixel values, the system extracts the minimum necessary data (endpoints defining the luminance range and weights defining the distribution), achieving compression efficiency while preserving color gamut and dynamic range in the reconstructed output.

The system changes the compression parameters by adapting the bit-depth allocation between endpoints and weights based on the actual HDR content characteristics. For images with limited luminance range or smooth gradients, the system uses fewer bits, improving compression efficiency. For images requiring full HDR capability, more bits are allocated, maintaining color gamut and dynamic range. This parameter adaptability resolves the contradiction between compression efficiency and HDR quality.