Scalable Video Encoding Weighted Prediction Brightness Variation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding methods, such as H.264/AVC and JSVC, face performance degradation due to non-linear brightness variations within slices or frames, particularly in images with shadows or flashes, as they rely on linear assumptions for weighting coefficients, leading to increased code requirements and decreased encoding efficiency.
Innovation Solution
A scalable video encoding method that computes weighting coefficients based on temporal brightness variations between decoded signals of reference and target frames, allowing for accurate weighted motion prediction without transmitting coefficient information, applicable to any unit image area, including macroblocks, and supporting both Explicit and Implicit modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If weighted motion prediction is performed using linear brightness variation assumptions, then encoding complexity is reduced, but prediction accuracy deteriorates for images with non-linear brightness changes
Solution Approach 1:
The patent changes the parameters used for weighting coefficient computation by incorporating multiple reference frames and their temporal distances, rather than relying on simple linear assumptions. This allows the system to adapt to non-linear brightness variations while maintaining manageable encoding complexity through standardized computation procedures.
Solution Approach 2:
The patent performs preliminary computation of weighting coefficients using temporal distance information from multiple reference frames before actual motion compensation. This preliminary action enables the system to pre-adjust for non-linear brightness changes, improving prediction accuracy without significantly increasing real-time encoding complexity.
2Measurement precision
If weighting coefficients are transmitted explicitly for each slice, then prediction accuracy is improved, but code amount increases
Solution Approach 1:
The patent uses the decoded signal from the immediately lower layer as a reference to compute weighting coefficients for the current layer. This copying approach allows the decoder to regenerate the same weighting coefficients as the encoder without transmitting them, thereby maintaining prediction accuracy while reducing code amount.
Solution Approach 2:
The system enables the decoder to self-generate weighting coefficients by utilizing the lower layer decoded signal and temporal distance information. This self-service mechanism eliminates the need for explicit coefficient transmission, reducing code overhead while maintaining the ability to handle non-linear brightness variations.
3Loss of information
If implicit mode weighted prediction is used for bi-predictive prediction only, then code amount is reduced, but adaptability to different prediction types deteriorates
Solution Approach 1:
The patent extends the implicit mode weighting coefficient computation method to work with various prediction types (P slices, B slices, L0/L1 prediction, bi-predictive prediction) rather than limiting it to bi-predictive prediction only. This universal approach allows the same coefficient generation mechanism to serve multiple prediction functions, improving adaptability without increasing code amount.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A scalable video encoding method of performing encoding by predicting an upper-layer signal having a relatively high spatial resolution by means of interpolation using an immediately-lower-layer signal having a relatively low spatial resolution. The method computes a first weighting coefficient for each image area of a predetermined unit size in a search for estimating a motion between an encoding target image area in an upper layer and a reference image area, where the first weighting coefficient is computed based on a brightness variation between an image area, which belongs to an immediately-lower layer and has the same spatial position as the encoding target image area, and the reference image area; and performs a motion estimation using a signal which is obtained by correcting a decoded signal of the reference image area by the first weighting coefficient and functions as an estimated signal in the motion estimation, so as to compute a motion vector. Then the method computes a second weighting coefficient based on a brightness variation between a reference image area indicated by the motion vector and the image area which belongs to the immediately-lower layer and has the same spatial position as the encoding target image area; and determines a signal, which is obtained by correcting a decoded signal of the reference image area by using the second weighting coefficient, to be a compensated signal in motion compensation, which functions as a predicted signal of the encoding target image area.