CABAC Binarization Reordering for Higher Codec Throughput
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The encoding and decoding throughput for Context-Based Adaptive Binary Arithmetic Coding (CABAC) in video coding systems is limited due to the complex derivation process of probabilities in the regular coding mode, especially when dealing with interleaved regular and bypass bins, which hinders efficient hardware-based acceleration.
Innovation Solution
Reordering the binarization results to collect bypass bins together for syntax elements like last_significant_coeff_x and last_significant_coeff_y, as well as for other syntax elements from prediction units and parameter groups, to facilitate improved encoding and decoding throughput by processing bypass bins sequentially.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If regular bins and bypass bins are interleaved in binarization results, then the coding can represent syntax elements accurately, but the encoding and decoding throughput is limited due to complex probability derivation in regular coding mode
Solution Approach 1:
The patent segments the binarization results by separating regular bins and bypass bins into distinct groups. By dividing the interleaved sequence into separate regular bin segments and bypass bin segments, the system can process each type independently, allowing parallel processing paths that improve throughput while managing complexity through structured organization of the coding operations
Solution Approach 2:
The patent reorders bins along a new dimension (processing order) rather than maintaining the original interleaved sequence. By changing the arrangement dimension from sequential interleaving to grouped ordering, the system enables more efficient hardware pipeline utilization and reduces the complexity of probability derivation by processing all regular bins followed by all bypass bins in a systematic sequence
2Productivity
If bypass bins are processed in interleaved positions with regular bins, then all syntax element information is captured, but hardware-based acceleration efficiency is reduced
Solution Approach 1:
The patent merges all bypass bins into a contiguous sequence separate from regular bins. By combining all bypass operations into a single continuous processing block, the system enables optimized hardware acceleration paths that can handle bypass bins efficiently without the overhead of interspersed regular bin processing, thereby improving both throughput and operational ease
Solution Approach 2:
The patent performs preliminary reordering of bins during the binarization stage to group bypass bins together before the actual coding process. This preliminary organization allows the hardware accelerator to receive pre-grouped bins that are optimized for acceleration, eliminating the need for complex runtime reordering and improving processing efficiency
Data Source
AI summary
A method and apparatus of context-based adaptive binary arithmetic encoding/decoding for syntax elements of one or more blocks are disclosed. According to one embodiment of the present invention, the binarization results of the x and y positions of the last significant coefficient of a transform unit are reordered by collecting the bypass bins together. According to another embodiment of the present invention, binarization results of syntax elements from two or more of four Intra N×N prediction units, including prev_intra_luma_pred_flag, rem_intra_luma_pred_mode, and mpm_idx are reordered by collecting the bypass bins together. In yet another embodiment according to the present invention, the binarization results of other syntax elements are also reordered by collecting bypass bins together. Furthermore, the above reordering of binarization results may be used for binarization results corresponding to a block and its neighboring blocks.