Video Decoder Prediction with Higher Precision Residuals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression techniques face challenges in achieving high compression efficiency while maintaining acceptable visual quality, as they often discard redundant information, leading to quality degradation and increased storage requirements for precise prediction data.
Innovation Solution
The method enhances video compression by using higher precision sample values during prediction, employing a video decoder with a predictor that operates at higher accuracy than the reconstructed image samples, and includes an adder for adding the predictor to differential samples, followed by down-scaling to the required accuracy, allowing for improved residual and differential residual reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If higher precision sample values are used during prediction, then compression efficiency is improved, but storage resources are increased
Solution Approach 1:
The patent segments the prediction process into two distinct stages: a high-precision prediction stage that operates on extended precision sample values to achieve accurate prediction, and a low-precision reconstruction stage that converts the prediction results to the required output precision. This segmentation allows the system to use high storage resources only temporarily during prediction while maintaining low storage requirements for the final reconstructed data.
Solution Approach 2:
The patent introduces a precision dimension as an additional layer in the signal processing pipeline. Instead of operating solely at the output precision level, the system adds a higher precision dimension for intermediate prediction operations, then projects the results back to the original precision dimension. This dimensional approach enables improved compression efficiency without permanently increasing storage requirements.
2Measurement precision
If higher precision prediction is applied, then rounding errors are minimized, but bit-rate increases
Solution Approach 1:
The patent applies preliminary high-precision prediction before the final reconstruction step. By performing prediction operations at extended precision first, the system minimizes rounding errors that would otherwise propagate through subsequent processing stages. The high-precision prediction results are then converted to the required output precision, ensuring accurate prediction while controlling the final bit-rate.
Solution Approach 2:
The patent dynamically changes the precision parameter during different stages of processing. The prediction stage operates with extended precision (higher parameter value), while the reconstruction stage uses the target precision. This parameter change allows the system to achieve high measurement precision during prediction without permanently increasing the bit-rate for storage and transmission.
3Reliability
If reconstructed samples are stored in higher accuracy, then prediction quality improves, but storage cost increases
Solution Approach 1:
The patent introduces high-precision prediction results as an intermediary element between the low-precision reconstructed samples and the final output. These intermediary prediction results operate at extended precision to ensure high reliability prediction quality, then are converted to the required precision for storage. This intermediary approach allows the system to benefit from high-precision prediction without incurring the full storage cost of permanently storing high-precision reconstructed samples.
Data Source
AI summary
A video decoder has a transform mode and a transform skip mode and in the transform skip mode a residual prediction mode is selected. A compressed bit-stream is entropy-decoded to obtain a block of quantised values. An inverse quantiser operates in the transform mode to form transform coefficients and in the transform skip mode to form differential residual values. In the transform mode an inverse transform is performed to form residual values. In the transform skip mode residual prediction is used perform residual prediction using high accuracy locally decoded residual value. The residual value is then down-scaled to obtain the reconstructed residual. A block predictor is formed from decoded image values at the same precision at which video samples are stored in the frame memory.