Directional Discrete Wavelet Transform for Video Edge Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video compression techniques, such as Mode-Dependent Directional Transforms (MDDT), face limitations in achieving high visual quality and accuracy, especially near feature edges, due to complexity, overhead, and the need for training sets, which negatively impact the reconstructed images in both intra and inter coding processes.
Innovation Solution
The implementation of Directional Discrete Wavelet Transforms (DDWT) that apply wavelet transforms along and then across feature edges, using differently sized transforms, such as 4×4, 8×8, and 16×16, without requiring a training set, to enhance visual quality and reduce complexity, thereby improving edge preservation and coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If Mode-Dependent Directional Transforms (MDDT) are used for video compression, then coding gain is improved, but device complexity increases and training overhead is required
Solution Approach 1:
The patent replaces complex trained transforms with simple, fixed wavelet transforms that require no training data or adaptive parameters. The DDWT uses standard wavelet bases (e.g., Daubechies, Coiflet) that are computationally inexpensive and universally applicable, eliminating the need for complex MDDT training infrastructure while maintaining compression effectiveness.
Solution Approach 2:
The patent changes the transform parameters from adaptive, data-dependent MDDT parameters to fixed wavelet transform parameters. By using predetermined wavelet bases and fixed decomposition levels, the system achieves consistent performance across different video content without requiring parameter adjustment or training, thereby reducing device complexity while preserving coding gain.
2Manufacturing precision
If MDDT is applied to improve visual quality, then edge preservation is enhanced, but overhead for detection and transmission increases
Solution Approach 1:
The patent extracts only the essential directional information needed for edge preservation by applying wavelet transforms along and across prediction directions. This eliminates the need for explicit edge detection and transmission of directional parameters, as the wavelet basis functions inherently capture edge orientations through their anisotropic filtering properties, thereby reducing overhead while maintaining edge preservation.
Solution Approach 2:
The patent makes the transform universally applicable to all prediction modes and edge orientations by using isotropic wavelet bases that can represent any direction. This multi-functional approach eliminates the need for mode-specific transform selection and parameter transmission, achieving edge preservation across all scenarios without increasing overhead, unlike MDDT which requires explicit directional information.
3Productivity
If traditional DCT transform is used for intra coding, then compression is achieved, but visual quality near feature edges deteriorates
Solution Approach 1:
The patent introduces asymmetry into the transform by applying wavelet decomposition with different orientations along and across prediction directions. This asymmetric treatment of directional frequencies allows the transform to adapt to edge orientations without requiring explicit edge detection, improving visual quality near edges while maintaining compression efficiency through directional sparsification of the transform coefficients.
Solution Approach 2:
The patent transitions from the traditional 2D separable DCT to a multi-dimensional wavelet transform that processes signals in multiple directional dimensions simultaneously. By decomposing the image into diagonal, horizontal, and vertical subbands, the DDWT captures edge information more effectively in all orientations, improving visual quality near edges while maintaining compression through the sparsity of wavelet coefficients in the transformed domain.
Data Source
AI summary
An apparatus and method for encoding video using directional discrete waveform transforms (DDWT), such as within a codec device. DDWT can be utilized to replace the use of intra transforms and inter transforms within the encoding system. In many ways the output of the DDWT can be compared with that provided using MDDT, however, it does not require a training process while it also provides enhanced encoding of feature edges with desirable visual characteristics. The transforms are applied in at least two passes, along the prediction direction, and then across the prediction direction, instead of being applied in fixed vertical and horizontal directions. Directional scaling is not required prior to the second stage of transforms.