Video Coding Transform Matrix Using Relaxed Orthogonality
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video compression systems face challenges in achieving real-time transmission of high-quality video over bandwidth-limited connections due to the inefficiencies in current integer transform methods, which often result in inaccuracies and complexity from real number representations.
Innovation Solution
The development of a method using transform matrix derived from DCT or KLT with basis vectors that are close to orthogonal, with deviations less than 1% in norms and elements less than 32, allowing for integer-based transforms that are close to optimal, yet relaxed in orthogonality and norm equality, to improve coding efficiency and reduce complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional integer transform methods are used, then complexity is reduced for real-time processing, but coding efficiency and picture quality deteriorate due to inaccuracies from real number representations
Solution Approach 1:
The patent changes the fundamental parameters of the transform matrix by using non-orthogonal basis vectors with controlled norm deviations (less than 1%) and small integer elements (less than 32). This parameter change enables the transform to maintain high coding efficiency while being implementable with integer arithmetic for real-time processing.
Solution Approach 2:
The patent employs simplified integer-based transform matrices that can be rapidly computed and discarded, replacing complex real-number transforms. These integer transforms provide sufficient accuracy for real-time applications without requiring precise floating-point arithmetic, effectively using 'cheaper' computational operations.
2Measurement precision
If basis vectors are made exactly orthogonal with equal norms, then mathematical precision is improved, but device complexity and computational burden increase
Solution Approach 1:
The patent applies partial orthogonality rather than exact orthogonality, allowing basis vectors to be 'close to orthogonal' with norm deviations less than 1%. This partial application of the orthogonality principle reduces computational complexity while maintaining sufficient precision for video coding applications.
Solution Approach 2:
The patent relaxes the strict parameters of exact orthogonality and equal norms, allowing small deviations in norm (less than 1%) while maintaining near-orthogonality. This parameter relaxation enables the use of simpler integer arithmetic while preserving the essential properties needed for effective transform coding.
3Loss of information
If transform coefficients are directly represented, then information accuracy is maintained, but bit capacity requirements become too high for bandwidth-limited transmissions
Solution Approach 1:
The patent transforms pixel values using a specially designed integer transform matrix that concentrates energy in fewer coefficients. The non-orthogonal basis vectors with controlled deviations create a transform domain where most information is captured in a small number of significant coefficients, enabling efficient quantization and compression.
Solution Approach 2:
The transform process extracts the most significant information components into a concentrated set of transform coefficients. By using the specific integer transform matrix, the patent extracts essential visual information that can be represented with fewer bits after quantization, separating important data from less critical details.
Data Source
AI summary
The present invention applies to video coding/decoding and discloses a method for transforming to/from transform coefficients and residual pixel data in moving pictures by a set of semi-ortonormal basis vectors. The basis vectors are derived from conventional DCT or KTL matrix'es, but relaxes to some extend the requirements for ortogonality, norm equality and element size limitation. In this way the present invention provides improved coding efficiency and lower complexity compared to previously used integer transforms.