Entropy Coding Adaptation for Video Prediction Gaps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing entropy coding methods in video compression systems are inefficient when no pixels are available for prediction, leading to increased bit usage due to unreasonably long VLC codes, particularly in situations where rapid and lasting light changes occur, compromising video quality.
Innovation Solution
The method involves monitoring for 'no prediction' situations and switching to a different entropy coding strategy where the DC coefficient is coded separately with special VLC tables, setting the prediction to a fixed value, and coding remaining coefficients with normal procedures, optimizing bit usage by using shorter code words for small numbers and longer code words for larger numbers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional entropy coding with prediction is used, then video quality is maintained under normal conditions, but bit usage increases significantly when no prediction data is available
Solution Approach 1:
The patent implements dynamic switching between two entropy coding modes (prediction mode and no-prediction mode) based on the availability of prediction data. The system adapts its coding strategy in real-time: when prediction data is available, it uses prediction-based coding for efficiency; when prediction data is unavailable, it switches to no-prediction mode with fixed VLC codes to maintain reliability while controlling bit usage.
Solution Approach 2:
The patent changes the coding parameters (VLC tables and coding strategy) based on the prediction data availability condition. In no-prediction mode, it uses specifically designed fixed VLC codes that are optimized for the absence of prediction data, rather than using the standard prediction-based VLC codes that would produce excessively long code words in this scenario.
2Ease of manufacture
If standard VLC codes are used without adaptation, then coding simplicity is maintained, but coding efficiency deteriorates in error-resilient scenarios
Solution Approach 1:
The patent segments the entropy coding process into two distinct pathways: prediction mode for normal operation and no-prediction mode for error-resilient scenarios. Each segment has its own optimized VLC tables and coding procedures, allowing the system to achieve high efficiency in both conditions without compromising simplicity within each segment.
Solution Approach 2:
The patent introduces a condition-checking mechanism that acts as an intermediary between the prediction process and the entropy coding process. This intermediary determines which coding mode to activate based on prediction data availability, seamlessly bridging the two coding strategies and ensuring optimal efficiency without requiring complex reconfiguration.
3Speed
If prediction data is used when unavailable, then coding speed is maintained, but bit usage increases and video quality deteriorates
Solution Approach 1:
The patent performs a preliminary check to determine the availability of prediction data before initiating the entropy coding process. This preliminary action prevents the system from attempting to use unavailable prediction data, thereby avoiding the generation of excessively long code words and ensuring that the appropriate coding mode is selected from the outset.
Solution Approach 2:
Instead of attempting to use prediction data and then handling the failure case, the patent inverts the approach by first checking for the absence of prediction data and directly activating the no-prediction mode when needed. This inversion eliminates the inefficiency of trying to use unavailable data and directly applies the optimized coding strategy for the actual condition.
Data Source
Figure 1~2
Figure 3~4
AI summary
The present invention provides a method for entropy coding of quantized transform coefficients when for some reason no pixels are available for prediction and the VLC codes which is based on statistics for available prediction data, is inexpediently long. According to the invention, it is determined whether a situation of no available prediction pixels occur. If so, a fixed value is inserted in a block of pixels which is used as the prediction block for deriving the residual block, which in turn are transformed and quantized. In the following entropy coding, a special purpose VLC is then used to represent the low frequency coefficients of the quantized transform coefficients.