Multiple sign bit hiding within a transform unit

What is this patent about?

’184 is related to the field of video data compression, specifically the efficient encoding and decoding of transform coefficients. In modern video standards like HEVC, residual data is converted into spectral coefficients, quantized, and then entropy encoded. A significant portion of the resulting bitstream is dedicated to specifying the sign (positive or negative) of these non-zero coefficients. Traditional methods often allocate one full bit per non-zero coefficient to indicate its sign, which represents a substantial overhead in high-bitrate or high-resolution video streams.

The underlying idea behind ’184 is to reduce bitstream overhead by embedding the sign of a specific coefficient within the mathematical properties of a group of coefficients, a technique known as sign bit hiding. Rather than explicitly signaling every sign bit, the invention partitions a transform unit into multiple non-overlapping subsets and uses the parity (even or odd nature) of the sum of the absolute values of the coefficients within each subset to represent the sign of one hidden coefficient. This allows the encoder to omit one sign bit per subset from the bitstream, provided the subset meets certain density or magnitude criteria.

The claims of ’184 focus on a method and apparatus for processing two or more distinct sets of sign bits within a single transform unit. For each set, the decoder is tasked with summing the absolute values of the reconstructed coefficients to derive a parity value, which then dictates the sign assigned to a specific coefficient in that set. On the encoding side, the claims cover the process of calculating this parity and, if it does not naturally match the sign of the coefficient to be hidden, adjusting a coefficient level by a single unit to flip the parity and ensure the hidden information is correctly embedded.

In practice, the invention leverages the existing modular structure of multi-level significance maps, where transform units are already divided into 4x4 coefficient groups. By applying a threshold test—such as the number of coefficients between the first and last non-zero values in a group—the system determines if a group is robust enough to hide a sign bit without introducing excessive distortion. If the threshold is met, the encoder uses a rate-distortion optimization or a simplified cost metric to decide which coefficient level to increment or decrement to achieve the required parity.

This approach differs from prior solutions that only attempted to hide a single sign bit for the entire transform unit. By enabling multiple hidden bits across partitioned subsets, ’184 significantly increases the potential compression gain, especially in larger 16x16 or 32x32 transform units. The integration with coefficient group scanning ensures that the sign bit hiding process aligns with the hardware-friendly modular processing used in advanced video codecs, providing a balance between computational complexity and bitstream efficiency.