Dynamic Image Encoding Motion Vector Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing moving picture encoding and decoding technologies, such as MPEG-4 AVC/H.264, face challenges in maintaining prediction accuracy and encoding efficiency, particularly in the time direct mode.
Innovation Solution
The proposed solution involves deriving multiple motion vector predictors from adjacent blocks and calculating the difference vector between the target block's motion vector and the selected predictor, which is then encoded to reduce the overall code amount.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion compensation prediction is used with multiple reference pictures, then prediction accuracy is improved, but coding information amount increases
Solution Approach 1:
The patent derives multiple motion vector predictors in advance from adjacent blocks before encoding the target block. By preparing these predictors beforehand, the system can select the most appropriate one without needing to encode multiple motion vectors, thus maintaining prediction accuracy while reducing coding information.
Solution Approach 2:
The patent changes the approach from encoding multiple motion vectors to encoding a difference value between the target motion vector and the selected predictor. This parameter transformation reduces the number of bits required to represent motion information while preserving prediction accuracy.
2Measurement precision
If multiple motion vectors are encoded for each block, then prediction accuracy is improved, but encoding complexity increases
Solution Approach 1:
The patent extracts only the essential motion information by deriving predictors from adjacent blocks and selecting the best match. Instead of encoding all possible motion vectors, the system extracts and encodes only the difference between the target vector and the selected predictor, simplifying the encoding process.
Solution Approach 2:
Motion vector predictors are derived in advance from neighboring blocks before the actual encoding of the target block. This preliminary derivation reduces the computational complexity during the main encoding process by eliminating the need to calculate and compare multiple motion vectors from scratch.
3Loss of information
If difference vector encoding is used, then code amount is reduced, but prediction accuracy may deteriorate
Solution Approach 1:
The patent uses a feedback mechanism where multiple motion vector predictors are derived from adjacent blocks and the best predictor is selected based on matching quality. This feedback loop ensures that the difference vector encoding maintains prediction accuracy by adaptively selecting the most appropriate predictor for each block.
Solution Approach 2:
The patent transforms the motion vector representation from absolute values to difference values relative to selected predictors. By changing the parameter representation and using adaptive predictor selection, the system reduces code amount while preserving prediction accuracy through intelligent parameter transformation.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
During inter-prediction, whereby a first block having each picture divided therein is divided into one or a plurality of second blocks, a spatial merge candidate generation unit (130) derives spatial merge candidates, without referring to blocks included in the first block, including the second block. A reference index derivation unit (131) for temporal merge candidates: sets reference index information for a temporal merge candidate as the value for reference index information for an encoded prediction block approaching the left side of a prediction block for encoding, when in a mode whereby the encoded block is divided into upper and lower prediction blocks at the horizontal interface; and sets reference index information for a temporal merge candidate as the value for reference index information for an encoded prediction block approaching the top side of the prediction block for encoding, when in a mode whereby the encoded block is divided into left and right prediction blocks at the vertical interface.