In-loop Filter Apparatus for Video Coding Using 2D Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional in-loop filters for video coding require complex rate-distortion optimization and large numbers of weighting coefficients, leading to increased computational complexity and inefficiency in suppressing quantization noise.
Innovation Solution
An improved in-loop filter apparatus that partitions reconstructed frames into 2D pixel blocks, applies a 2D transform, multiplies spectral components with gain coefficients dependent on neighboring components and filtering parameters, and uses inverse transforms to generate filtered frames, reducing the need for complex signaling and multiplication operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional in-loop filters use weighted sum of multiple pixels with transmitted weighting coefficients, then filtering quality improves, but device complexity and computational complexity increase significantly
Solution Approach 1:
The patent transforms the filtering operation from spatial domain to frequency domain by applying 2D DCT transform. The filtering is performed by multiplying DCT coefficients with gain factors derived from spectral components, avoiding the need for transmitting multiple weighting coefficients per block. This parameter transformation resolves the contradiction by maintaining filtering quality while dramatically reducing device complexity.
Solution Approach 2:
The patent replaces the mechanical multiplication operation (pixel values multiplied by weighting coefficients) with a spectral domain operation. Instead of performing weighted sums in spatial domain, the system uses DCT transform followed by coefficient multiplication in frequency domain, which is computationally more efficient and reduces the number of operations required.
2Manufacturing precision
If conventional in-loop filters transmit large numbers of weighting coefficients from encoder to decoder, then filtering accuracy improves, but loss of information increases due to signaling overhead
Solution Approach 1:
The patent changes the representation parameters from spatial domain weighting coefficients to frequency domain gain factors. The gain factors are derived from the spectral components of the block itself, requiring minimal signaling. This parameter transformation maintains filtering accuracy while minimizing information loss during transmission.
Solution Approach 2:
The patent uses the spectral components of the block itself to generate the filtering parameters, rather than transmitting separate weighting coefficients. This self-referential approach allows the decoder to reconstruct the filtering parameters from the received spectral data, eliminating the need for additional signaling and preventing information loss.
3Adaptability or versatility
If conventional in-loop filters use universal multipliers reloaded for each 2x2 pixel block, then filtering versatility improves, but productivity decreases due to repeated multiplication operations
Solution Approach 1:
The patent replaces repeated multiplication operations with DCT transform and coefficient multiplication in frequency domain. The DCT transform can be implemented using lookup tables and addition/subtraction operations, avoiding the need for universal multipliers. This substitution maintains filtering versatility while significantly improving processing productivity.
Solution Approach 2:
The patent performs the DCT transform as a preliminary step before filtering. The transform coefficients are computed once and then used for filtering operations, avoiding repeated multiplication operations for each 2x2 block. This preliminary action improves productivity by eliminating redundant computations while maintaining versatility through the flexibility of DCT-based filtering.
4Loss of information
If conventional in-loop filters perform rate-distortion optimization to reduce weighting coefficients, then loss of information decreases, but device complexity increases due to complex RDO procedures
Solution Approach 1:
The patent changes the optimization parameters from spatial domain weighting coefficients to frequency domain gain factors. The RDO can be performed more efficiently in the frequency domain because the gain factors are derived from spectral components, requiring less complex optimization procedures. This parameter transformation reduces both information loss and device complexity simultaneously.
Data Source
AI summary
The disclosure relates to an in-loop filter apparatus for video coding, which is configured for processing a reconstructed frame corresponding to a current frame for generation of a filtered reconstructed frame, wherein the reconstructed frame comprises a plurality of pixels, each pixel being associated with a pixel value. The in-loop filter apparatus comprises a processing unit configured to: partition the reconstructed frame into a plurality of overlapping and/or non-overlapping 2D pixel blocks; generate for each 2D pixel block a 2D spectrum by applying a 2D transform to the 2D pixel block, wherein the 2D spectrum comprises a plurality of spectral components; generate for each 2D pixel block a filtered 2D spectrum by multiplying each spectral component with a respective gain coefficient, wherein the respective gain coefficient depends on the respective spectral component and/or one or more neighboring spectral components of the respective spectral component and one or more filtering parameters; generate for each 2D pixel block a filtered 2D pixel block by applying an inverse 2D transform to the filtered 2D spectrum; and generate the filtered reconstructed frame on the basis of the plurality of filtered 2D pixel blocks.