Video Encoding Scale Angle Variation Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding technologies do not effectively utilize scale and angle variation information to reduce redundancy and improve compression efficiency, particularly in inter prediction methods.
Innovation Solution
The proposed solution involves determining regions in an image with scale and angle changes relative to a reference image, generating variation information for these changes, and encoding residual information to predict and encode the current image efficiently, using this information to minimize rate-distortion cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional inter prediction technology is used to generate predicted blocks, then video compression can be achieved, but unnecessary redundancy remains and compression efficiency is limited
Solution Approach 1:
The patent applies parameter changes by introducing scale variation and angle variation as additional transformation parameters beyond conventional motion vectors. The encoder determines scale variation information representing size changes and angle variation information representing rotational changes between reference and current blocks, thereby capturing object transformations more accurately and reducing residual redundancy.
Solution Approach 2:
The patent extends the conventional 2D motion compensation by adding scale and angle dimensions. Instead of only translating blocks based on motion vectors, the method transforms blocks across multiple dimensions including size scaling and rotational angle changes, enabling more comprehensive representation of object variations in video sequences.
2Measurement precision
If scale and angle variation information is generated and encoded, then prediction accuracy improves, but encoding complexity increases
Solution Approach 1:
The patent segments the complexity by dividing the encoding process into distinct modules: motion vector determination, scale variation determination, angle variation determination, and residual encoding. Each module handles a specific aspect of transformation, allowing independent optimization and simplifying the overall complex encoding process while maintaining high prediction accuracy.
3Adaptability or versatility
If motion vectors are used to represent position variations, then inter prediction can be performed, but scale and angle variations are not captured
Solution Approach 1:
The patent creates a universal prediction framework that incorporates multiple functions: conventional motion vector-based translation, scale variation-based resizing, and angle variation-based rotation. This multi-functional approach allows the system to adapt to various types of object transformations (positioning, scaling, rotating) within a unified prediction model, significantly improving prediction adaptability across different video scenarios.
Data Source
AI summary
A video encoding and decoding methods and apparatuses for performing the methods thereof are provided. The video encoding method includes: determining a region in a current image having a change in at least one of a scale and an angle with respect to a reference image; generating variation information representing a variation in at least one of the scale and the angle between the region in the current image and a reference region in the reference image corresponding to the region in the current image; and encoding residual information representing a difference between a predicted sample value from the reference image and a sample value of the current image based on the generated variation information.