Symmetrical FIR Resampling Filter Edge Handling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image resampling systems face challenges in converting between different sampling structures without introducing distortion, particularly in television formats, where additional samples outside the image area are required for reversible up- and down-conversion, which can exceed processing capacity in real-time video processing.

Innovation Solution

A method using symmetrical FIR resampling filters that copy values from inside the image to positions equidistant from the image edge to obtain contributions from outside the image, allowing for reversible resampling without the need to store additional samples, by modifying filter coefficients and reflecting samples to derive values from within the image.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If reversible FIR resampling filters are used for up- and down-conversion between different sampling structures, then image quality and reversibility are maintained, but additional samples outside the image area must be stored and processed, exceeding real-time processing capacity

Engineering Contradiction:
Improvereversibility of resamplingVSAvoidreal-time processing capacity
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies the copying principle by creating mirror images of the input image at both edges. These copied samples are then used as additional input samples for the FIR resampling filter, eliminating the need to store separate additional samples outside the image area. The mirror image copies provide the necessary filter inputs while maintaining reversibility and staying within real-time processing capacity.

Inventive Principle:
Principle #26Copying

2Manufacturing precision

If a large filter aperture is used to achieve accurate resampling conversion, then conversion accuracy is improved, but the number of additional samples required outside the image area increases

Engineering Contradiction:
Improveconversion accuracyVSAvoidnumber of additional samples
Core Design Contradiction:
Manufacturing precisionVSQuantity of substance

Solution Approach 1:

By creating mirror image copies of the input image, the patent provides additional sample values that extend the effective filter aperture without requiring physical storage of separate additional samples. The copied samples are generated on-the-fly and used immediately by the FIR filter, maintaining conversion accuracy while minimizing storage requirements.

Inventive Principle:
Principle #26Copying

3Measurement precision

If additional samples outside the image area are stored for reversible resampling, then resampling accuracy is maintained, but storage requirements and processing complexity increase

Engineering Contradiction:
Improveresampling accuracyVSAvoidstorage and processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent eliminates the need for separate storage of additional samples by generating mirror image copies of the input image. These copies are created and used in-place, reducing storage requirements while maintaining the accuracy needed for reversible resampling. The processing complexity is reduced because the copied samples are generated algorithmically rather than requiring separate storage and retrieval operations.

Inventive Principle:
Principle #26Copying

Data Source

PatentUS9047683B2Method and apparatus for resampling a spatially sampled attribute of an image
Publication Date: 2015.06.02 GRASS VALLEY LTD
  • US9047683B2 patent drawing
  • US9047683B2 patent drawing
  • US9047683B2 patent drawing

AI summary

Resampling a spatially sampled attribute of an image, for example up- or down-conversion, in which contributions to a symmetrical FIR resampling filter from positions outside the image—that is to say where the filter aperture extends beyond an image edge—are obtained by copying values from positions inside the image. This can conveniently be done by modifying filter coefficients.