Multiple sign bit hiding within a transform unit

What is this patent about?

’028 is related to the field of video data compression, specifically the efficient encoding and decoding of transform coefficients. In modern video codecs like HEVC, residual data is converted into the frequency domain via transforms, resulting in quantized coefficients. A significant portion of the resulting bitstream is dedicated to signaling the sign (positive or negative) of these coefficients. Traditional methods allocate one full bit per non-zero coefficient, which creates a data overhead that the invention seeks to reduce through a technique known as sign bit hiding.

The underlying idea behind ’028 is that the sign of a coefficient can be mathematically embedded into the magnitude of a group of coefficients, effectively signaling the sign for free. By treating the parity (even or odd status) of the sum of absolute values within a specific set of coefficients as a proxy for a sign bit, the encoder can communicate a sign without an explicit flag. If the natural parity of the quantized levels does not match the required sign, the encoder slightly adjusts one coefficient's magnitude to flip the parity, utilizing a threshold test to ensure that such adjustments do not disproportionately degrade image quality.

The claims of ’028 focus on a method and apparatus for reconstructing coefficients by partitioning a transform unit into multiple coefficient groups and applying sign bit hiding selectively to each group. The process involves determining if a specific group qualifies for sign bit hiding by checking if the distance (count of coefficients) between the first and last non-zero coefficients in that group exceeds a predefined threshold. For qualifying groups, the decoder calculates the sum of absolute values and assigns a sign to a syntax element based on whether that sum is even or odd.

In practice, the invention operates by leveraging the modular structure of multi-level significance maps. By dividing a large transform unit into smaller 4x4 sub-blocks, the system can hide multiple sign bits across a single transform unit rather than just one. The encoder performs a cost-benefit analysis, often using a simplified rate-distortion metric, to decide which coefficient level to increment or decrement by one to satisfy the parity requirement. This ensures that the hidden bit is recovered perfectly by the decoder simply by summing the reconstructed magnitudes.

This approach differs from prior methods by moving beyond global sign hiding at the transform unit level to a more granular, group-based implementation. By tying the hiding mechanism to the spatial distribution of non-zero coefficients (the distance between the first and last non-zero values), the invention avoids the overhead of hiding bits in sparse groups where the distortion cost of adjusting a level would outweigh the savings of the hidden bit. This selective, multi-level application allows for higher compression efficiency in high-resolution video where transform units are large and contain numerous coefficients.