Inter-frame Prediction Coding Using Template Matching and Offset Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current inter-frame motion compensation prediction methods in video coding, such as H.264/AVC, face inefficiencies due to the need to encode motion vector information explicitly, leading to increased bitstream size and computational complexity, especially when using arithmetic averaging for multi-hypothesis prediction which results in low prediction performance and coding efficiency.
Innovation Solution
The method calculates distortions between a template region of the current coding block and multiple matching templates in reference frames to determine offset vectors, obtaining hypothesis prediction values and calculating a template matching prediction value through weighted averaging, which reduces the need to encode offset vector information and simplifies computations by performing calculations in integer precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If motion vector information is explicitly encoded in the code stream, then the decoder can perform motion compensation prediction, but the bitstream size increases
Solution Approach 1:
The patent extracts only the essential motion information (offset vectors) needed for prediction while removing redundant explicit motion vector encoding from the bitstream. The encoder calculates multiple hypothesis prediction values using template matching and offset vectors, then transmits only the residual and necessary offset information, allowing the decoder to reconstruct predictions without receiving full motion vectors.
Solution Approach 2:
The patent creates multiple copies of prediction hypotheses from different reference frames and templates. Instead of encoding a single motion vector, the system generates multiple hypothesis prediction values by copying and adapting template regions from different reference frames, then selects or combines these hypotheses to form the final prediction, reducing the need for explicit motion vector transmission.
2Reliability
If arithmetic averaging is used for multi-hypothesis prediction, then prediction values can be obtained, but computational complexity increases due to floating point operations
Solution Approach 1:
The patent changes the parameter of averaging operation from arithmetic mean (floating point) to geometric mean or other alternative operations. Specifically, it uses the formula: prediction value = round(sqrt(hypothesis1 × hypothesis2)) or similar geometric-based operations, which can be implemented using integer arithmetic and bit shifts, significantly reducing computational complexity while maintaining prediction accuracy.
Solution Approach 2:
The patent replaces the mechanical floating-point arithmetic system with an integer-based computational system. By using geometric mean operations, bit shifts, and integer multiplication/division, the system substitutes complex floating-point hardware requirements with simpler integer arithmetic units, reducing device complexity and power consumption while achieving similar prediction performance.
3Reliability
If multiple hypothesis predictions from one reference frame are averaged, then prediction values are obtained, but both prediction performance and coding efficiency remain low
Solution Approach 1:
The patent extends the prediction search from a single reference frame to multiple reference frames (temporal dimension). Instead of averaging multiple hypotheses from one frame, the system searches for matching templates across L different reference frames, obtaining offset vectors and hypothesis predictions from multiple temporal dimensions. This multi-frame approach provides more diverse and accurate prediction hypotheses, improving both prediction performance and coding efficiency by better exploiting temporal redundancy.
Data Source
AI summary
Inter-frame prediction coding method, device and system are provided. The inter-frame prediction coding method includes: calculating distortions between a template area of current encoding block and each of M matching templates in L reference frames, to determine M offset vectors; acquiring M hypothesis prediction values of the encoding block to which the M matching templates correspond according to the determined M offset vectors, and calculating the template matching prediction value of the current encoding block according to the M hypothesis prediction values; comparing the template matching prediction value and original value of the current encoding block to acquire the residual of the current encoding block, and encoding the residual. The technical solution improves prediction performance of the video coding system and increases coding efficiency.