Interpolation Filters for Adaptive Image Block Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image coding methods require storing multiple versions of reference images at different sizes in memory, leading to increased memory usage, which is particularly problematic in hardware implementations with limited surface area.
Innovation Solution
A method that uses separable interpolation filters to jointly perform upsampling or subsampling and motion compensation, depending on the size of the reference and current images, eliminating the need to store additional upsampled or subsampled reference images in memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple versions of reference images at different sizes are stored in memory to support adaptive resolution, then the system can perform motion compensation between images of different sizes, but the memory usage increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-defining a set of interpolation filters that can handle different size relationships between reference and current images. These filters are prepared in advance and selected based on the specific size relationship, avoiding the need to store multiple pre-processed image versions.
Solution Approach 2:
The patent changes the parameter of image size handling by using interpolation filters with different characteristics depending on whether the reference image is smaller or larger than the current image. The system selects appropriate filters based on size parameters, transforming the reference image to the current image size dynamically during processing rather than storing multiple versions.
2Loss of energy
If the quantization step is greatly increased to reduce output bitrate at low bitrates, then the bitrate is reduced, but the distortion of the image signal increases significantly
Solution Approach 1:
The patent segments the image processing into different resolution levels by supporting adaptive resolution. Instead of uniformly quantizing the entire image at high levels, the system can work with subsampled versions of the image at appropriate resolution levels, reducing the overall quantization burden and maintaining better image quality at low bitrates.
Solution Approach 2:
The patent introduces another dimension of resolution adaptation by allowing the system to switch between different image sizes. This dimensional flexibility enables the system to maintain acceptable image quality at lower bitrates by appropriately matching the resolution level to the available bitrate, rather than relying solely on quantization adjustments.
3Adaptability or versatility
If reference images are upsampled or subsampled to match the size of the current image before motion compensation, then motion compensation can be performed between images of different sizes, but additional processing steps and computational complexity are required
Solution Approach 1:
The patent applies universality by creating a unified motion compensation framework that handles both same-size and different-size image pairs using a common set of interpolation filters. The system universally applies the appropriate filter based on size comparison, eliminating the need for separate processing paths and reducing overall system complexity.
Solution Approach 2:
The patent introduces dynamics by making the interpolation filter selection adaptive and dynamic based on the size relationship between reference and current images. The system dynamically selects the appropriate filter (for upsampling or downsampling) based on real-time size comparisons, allowing flexible handling of various size relationships without requiring static pre-processing of all possible size combinations.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A method is described for reconstructing a current block of a current image from at least one block of a reconstructed reference image at a different size from the size of the current image. The method comprises the steps: - motion compensate (10) the reference block with at least one interpolation filter that depends on the size of said reference image and on the size of said current image, - decode (12), for the current block, a residue block, and - reconstruct (14) the current block from the residue block and from the motion compensated reference block.