Multiple sign bit hiding within a transform unit

What is this patent about?

’622 is related to the field of video data compression, specifically the efficient encoding and decoding of transform coefficients within a video codec. In modern standards like HEVC, residual data is transformed into the frequency domain and quantized, resulting in a sparse matrix of coefficients. A significant portion of the resulting bitstream is dedicated to signaling the sign (positive or negative) of these non-zero coefficients. The patent addresses the overhead of these sign bits by employing data hiding techniques that embed sign information within the mathematical properties of the coefficients themselves.

The underlying idea behind ’622 is to utilize the parity of a group of quantized coefficients to signal the sign of one specific coefficient in that group, thereby eliminating the need to explicitly transmit its sign bit. By summing the absolute values of coefficients within a defined set, the encoder can ensure the resulting sum is even or odd to match the sign of a hidden bit. If the natural parity does not match the required sign, the encoder makes a subtle, rate-distortion-optimized adjustment to one coefficient's level to flip the parity, allowing the decoder to infer the missing sign through a simple summation.

The claims of ’622 focus on a method and apparatus for reconstructing coefficients where the transform unit is partitioned into non-overlapping coefficient groups, and the decision to apply sign bit hiding is made dynamically for each group. Specifically, the claims require determining whether to use sign bit hiding for a current group based on whether an adjacent coefficient group contains at least one non-zero coefficient. When hiding is active, the decoder reconstructs all but one sign bit from the bitstream and derives the final sign by calculating whether the sum of absolute values in that group is even or odd.

In practice, the invention leverages the existing modular structure of multi-level significance maps, where transform units are already divided into 4x4 sub-blocks. The system evaluates these sub-blocks against a threshold—such as the number of non-zero coefficients or the distance between them—to ensure that the cost of adjusting a coefficient level to satisfy parity does not outweigh the savings of hiding the sign bit. This dynamic grouping approach allows the codec to hide multiple sign bits within a single large transform unit, scaling the efficiency gains with the complexity of the video data.

This approach differs from prior sign-hiding methods that typically operated on an entire transform unit or only hid the sign of the very first non-zero coefficient. By localizing the hiding process to specific coefficient groups and using neighboring group occupancy as a trigger, the invention provides a more granular and adaptive mechanism. It ensures that sign bit hiding is only applied where there is sufficient 'budget' in the coefficients to absorb a parity adjustment without significant quality loss, while maximizing the total number of bits saved across high-resolution frames.