Method for sub-pixel value interpolation

What is this patent about?

’469 is related to the field of digital video coding and decoding, specifically focusing on motion-compensated temporal prediction. In modern video compression, motion vectors often point to sub-pixel locations to better model real-world movement. This requires interpolating values for pixels that do not exist in the original sampled grid, a process that traditionally forces a trade-off between high computational complexity and heavy memory consumption.

The underlying idea behind ’469 is a multi-stage interpolation strategy that balances precision and efficiency by using diagonal linear interpolation for the finest sub-pixel resolutions. Instead of deriving every fractional pixel through complex multi-tap filtering or relying on previously truncated values that lose accuracy, the invention calculates primary sub-pixels using weighted sums and then determines the remaining fractional positions through simple averages of these primary points. This approach maintains high visual quality while significantly reducing the mathematical burden on the decoder.

The claims of ’469 focus on a method for generating sub-pixel values at fractional locations defined by a power of two. The process involves directly interpolating half-unit horizontal and vertical positions using weighted sums of original pixels. For the most granular fractional locations, such as quarter-pixel or eighth-pixel positions, the claims specify taking a weighted average of two existing pixels or sub-pixels that are located diagonally relative to the target sub-pixel, ensuring a streamlined calculation path.

In practice, the system implements a hybrid of before-hand and on-demand interpolation. The encoder pre-calculates a subset of sub-pixels to save time during the exhaustive motion estimation search, while the decoder only calculates the specific sub-pixel values indicated by the received motion vectors. By using intermediate values with a dynamic range greater than the final pixel depth, the invention avoids the rounding errors that typically plague multi-stage interpolation schemes, ensuring the encoder and decoder remain perfectly synchronized.

This invention differs from prior approaches like TML5 and TML6 by eliminating the need for high-precision arithmetic across all sub-pixel calculations. Unlike TML5, which suffers from precision loss due to cumulative clipping, ’469 uses linear interpolation from primary sub-pixels to maintain accuracy. Compared to TML6, it drastically reduces memory requirements because it does not need to store massive arrays of high-bit-depth intermediate values, making it ideal for implementation in resource-constrained mobile telecommunications terminals.