Intra Block Copy Decoding With Template-Matched BV Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Intra Block Copy (IBC) encoding in video coding suffers from low prediction accuracy due to insufficient consideration of deviations in reconstructed samples, leading to inefficiencies in video encoding and decoding processes.
Innovation Solution
The proposed method employs template matching and Block Vector Prediction (BVP) techniques to refine the Block Vector (BV) selection process, utilizing cost functions like SAD, SATD, MSE, SSD, MAD, and MSD, and incorporates advanced motion vector prediction methods such as AMVP and ABVP to enhance prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If reconstructed samples in a best matching reference block are directly used as prediction samples, then the encoding process is simple, but prediction accuracy is low due to significant deviations in some scenarios
Solution Approach 1:
The patent applies preliminary action by performing sample offset compensation before using the reference block samples as prediction samples. The encoder calculates offset values based on motion vectors and applies these offsets to the reference samples in advance, so that the compensated samples are ready for prediction. This preliminary compensation step improves prediction accuracy without significantly complicating the encoding process.
Solution Approach 2:
The patent changes the parameter of the prediction samples by applying sample offset compensation. Instead of directly using the original reference block samples, the encoder modifies the sample values by adding calculated offset values. This parameter change allows the prediction samples to better match the current block, reducing deviations and improving prediction accuracy in various scenarios.
2Measurement precision
If advanced motion vector prediction methods such as AMVP and ABVP are incorporated, then prediction accuracy is enhanced, but device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the motion vector prediction process into distinct stages: generating candidate motion vectors, selecting the best candidate using cost functions, and applying sample offset compensation. This segmentation allows each stage to be optimized independently, managing complexity while maintaining high prediction accuracy through systematic processing.
Solution Approach 2:
The patent uses feedback mechanisms by evaluating multiple motion vector candidates using cost functions (SAD, SATD, MSE, SSD, MAD, MSD) and selecting the best candidate based on the evaluation results. This feedback-driven selection process ensures high prediction accuracy while managing complexity through systematic comparison and selection rather than exhaustive search.
3Measurement precision
If template matching and Block Vector Prediction techniques are employed to refine BV selection, then prediction accuracy is improved, but encoding time increases
Solution Approach 1:
The patent applies partial action by performing template matching and sample offset compensation only when beneficial, rather than for all blocks. The encoder uses cost functions to evaluate whether refined BV selection will improve prediction accuracy, and only applies the additional processing when the evaluation indicates it will be worthwhile. This selective approach improves accuracy for difficult cases while minimizing unnecessary encoding time for simpler cases.
Solution Approach 2:
The patent performs preliminary actions by generating candidate block vectors and evaluating them using cost functions before finalizing the BV selection. This preliminary evaluation allows the encoder to identify promising candidates that warrant further refinement through template matching and offset compensation, while discarding poor candidates early to save encoding time.
Data Source
AI summary
A method for decoding includes: a first template of a current block is determined, and a BV candidate list of the current block is determined; a bitstream is decoded to determine BV indication information, and determine a first BV according to the BV indication information and the BV candidate list; a matching template and a reference block corresponding to the matching template are determined according to the first BV; model parameters are determined according to the first template and the matching template, the reference block is filtered according to the model parameters, and a filtered reference block is determined; and a prediction value of the current block is determined according to the filtered reference block.