Adaptive Transform Encoding Reducing Bitrate Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video and picture codecs face a significant bitrate overhead when encoding content-adaptive transforms, as the size of the transform increases rapidly with the size of the block, leading to increased encoding costs compared to the bit-rate of the sequence, especially at low bit rates.
Innovation Solution
The method involves determining and encoding the precision of transform vectors, allowing only the most significant vectors to be encoded and the rest generated or set to zero, reducing the overhead while maintaining performance by using a block-based adaptive transform learning algorithm and scalar quantization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If content-adaptive transforms are used to improve energy compaction and reduce bitrate, then transformation efficiency is improved, but the bitrate overhead for encoding the transforms increases significantly
Solution Approach 1:
The transform is divided into multiple transform vectors (rows and columns). Instead of encoding the complete transform matrix, only a selected subset of transform vectors is encoded and transmitted. The remaining vectors are derived or reconstructed at the decoder side, thereby segmenting the encoding task to reduce bitrate overhead while preserving transformation efficiency.
Solution Approach 2:
The essential components (selected transform vectors) are extracted from the complete transform and transmitted to the decoder. The non-essential components (remaining vectors) are not transmitted but reconstructed at the decoder using derivation relationships, thus extracting only the necessary information to reduce overhead.
2Measurement precision
If the size of the block increases to improve resolution, then picture quality is improved, but the encoding cost increases rapidly as N^4
Solution Approach 1:
The large transform matrix is segmented into multiple smaller transform vectors. By encoding only selected vectors and deriving others at the decoder, the complexity of encoding large transforms is reduced from O(N^4) to a much lower complexity operation, enabling high-resolution blocks to be processed efficiently.
Solution Approach 2:
Instead of encoding all transform vectors completely, only a partial set of vectors (the most significant ones) is encoded and transmitted. The remaining vectors are reconstructed at the decoder, applying partial action to reduce encoding cost while maintaining sufficient picture quality.
Data Source
AI summary
A method for decoding a bitstream representative of a picture is disclosed. Information representative of at least one encoding precision is first determined. A transform is then decoded responsive to the determined information. Finally, the picture is decoded using the decoded transform. A method for encoding a bitstream representative of a picture, as well as encoding devices and decoding devices are also disclosed.