Adaptive Coefficient Binning for Efficient Transform Entropy Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional transform coding techniques are inefficient for compressing wide dynamic range transform coefficients, as they lead to large dynamic range transform coefficients with a Laplacian distribution, resulting in inefficient entropy coding due to frequent zero symbols and increased symbol alphabet, which is not well-suited for variable length entropy coding.
Innovation Solution
A digital media codec represents wide dynamic range transform coefficients in two parts: a normalized coefficient and a bin address, where the normalized coefficient is encoded using variable length entropy coding and the bin address is encoded using fixed length coding, with adaptive bin size adjustment based on the count of non-zero coefficients to resemble a narrow range distribution, allowing for efficient entropy coding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional transform coding is used on wide dynamic range input data, then transform coefficients are generated, but the coefficients have a wide dynamic range with Laplacian distribution, making entropy coding inefficient
Solution Approach 1:
The transform coefficient is segmented into two parts: a normalized coefficient (higher-order bits) and a bin address (lower-order bits). This segmentation allows the normalized coefficient to be entropy coded efficiently while the bin address uses fixed-length coding, resolving the inefficiency of directly entropy coding the entire wide-range coefficient.
Solution Approach 2:
The patent changes the parameter representation by normalizing the transform coefficient through division by a scale factor, transforming the wide dynamic range coefficient into a normalized coefficient with a more favorable distribution for entropy coding. This parameter transformation converts the Laplacian distribution into a distribution better suited for variable-length entropy coding.
2Loss of information
If wide dynamic range transform coefficients are directly entropy coded, then compression is attempted, but the symbol alphabet becomes large and frequent zero symbols reduce coding efficiency
Solution Approach 1:
By segmenting the coefficient into normalized part and bin address, the patent reduces the symbol alphabet size for entropy coding. The normalized coefficient has a smaller range and more favorable distribution, reducing the number of unique symbols that need to be entropy coded, thereby lowering device complexity while maintaining compression efficiency.
Solution Approach 2:
The patent extracts the lower-order bits (bin address) from the transform coefficient and handles them separately with fixed-length coding. This extraction removes the problematic frequent zero symbols from the entropy coding process, as the bin address explicitly represents the offset from the normalized coefficient, eliminating the need to encode frequent zero patterns.
3Productivity
If adaptive bin size adjustment is implemented, then narrow range distribution is achieved for better entropy coding, but additional processing is required
Solution Approach 1:
The patent implements dynamic adaptation by adjusting the bin size (scaling factor) based on the statistical properties of the transform coefficients. The bin size is adapted to match the local characteristics of the coefficient distribution, creating a dynamic coding scheme that optimizes entropy coding efficiency for different regions of the frequency spectrum while managing complexity through localized adaptation.
Solution Approach 2:
The adaptive bin size adjustment uses feedback from the coefficient statistics to optimize the normalization process. By monitoring the distribution characteristics and adjusting the bin size accordingly, the system creates a feedback loop that continuously optimizes the entropy coding efficiency without requiring complex global optimization, thereby managing device complexity while improving productivity.
Data Source
AI summary
A block transform-based digital media codec more efficiently encodes wide dynamic range transform coefficients in two parts: a normalized coefficient and bin address. The normalized coefficient relates to a grouping of coefficient values of the wide dynamic range into bins, whereas the bin address is an index of the coefficient value within a bin. With careful selection of the bin size, the normalized coefficients have a probability distribution more similar to narrow range transform coefficients, which is better suited to variable length entropy coding. The codec uses variable length entropy coding to encode the normalized coefficients in a “core” of the compressed bitstream, and fixed length coding to encode the bin address as a separate optional layer that can be omitted. The codec further adaptively varies the bin size of the grouping based on a backward adaptation process to adjust the normalized coefficients toward a probability distribution well suited for efficient variable length entropy coding.