Image Upsampling via Sparse Approximation for Artifact Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image upsampling methods often result in blurring, ringing, and aliasing artifacts, especially when dealing with sharp edges, and fail to maintain temporal coherence in video playback, due to their reliance on band-limited assumptions and linear interpolation models.

Innovation Solution

The proposed solution combines an initial upsampling process with a sparse approximation technique, using high-frequency portions extracted from the signal to refine the upsampling result, thereby enhancing image quality by improving high-frequency components without relying on the downsampling process or specific edge detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If conventional linear interpolation methods (bilinear, bicubic) are used for upsampling, then computational complexity is low and implementation is simple, but image quality deteriorates with blurring, ringing, and aliasing artifacts

Engineering Contradiction:
Improveimplementation simplicityVSAvoidimage quality
Core Design Contradiction:
Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent segments the upsampling process into two distinct stages: first applying a simple linear interpolation method to generate an initial upsampled image, then separately extracting and enhancing high-frequency components through sparse approximation. This segmentation allows each stage to be optimized independently - the first stage maintains computational simplicity while the second stage recovers lost high-frequency details, thereby resolving the contradiction between implementation ease and image quality.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces sparse approximation as an intermediary technique between the initial upsampling and the final image enhancement. This intermediary step extracts high-frequency components that were lost during conventional linear interpolation, acting as a bridge that transfers the simplicity advantage of linear methods while adding the quality enhancement of advanced signal processing. The intermediary sparse approximation step enables the system to achieve both low computational complexity and high image quality.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Manufacturing precision

If linear models with higher degree basis functions are used to capture higher order statistics, then edge representation improves, but support in spatial domain increases causing ringing artifacts around edges

Engineering Contradiction:
Improveedge representation accuracyVSAvoidringing artifacts
Core Design Contradiction:
Manufacturing precisionVSObject-generated harmful factors

Solution Approach 1:

The patent extracts high-frequency components separately from the low-frequency base image using sparse approximation. By taking out the high-frequency details and adding them back to the upsampled image, the method achieves accurate edge representation without using high-degree basis functions that would cause ringing artifacts. This extraction approach allows the system to capture edge statistics without the spatial support expansion problem.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter representation by using sparse approximation in a transform domain rather than directly manipulating spatial domain basis function degrees. This parameter transformation allows the system to capture higher-order statistics and improve edge representation while avoiding the ringing artifacts that result from increasing the degree of spatial basis functions. The parameter change from spatial domain to transform domain resolves the contradiction.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If directional interpolation schemes like NEDI are used to perform interpolation along edge directions, then local covariance utilization improves edge handling, but artifacts remain in the resulting images

Engineering Contradiction:
Improveedge direction accuracyVSAvoidinterpolation artifacts
Core Design Contradiction:
Manufacturing precisionVSObject-generated harmful factors

Solution Approach 1:

The patent merges two different approaches: the simple linear interpolation method that provides a smooth base image, and the sparse approximation method that recovers high-frequency details including edge information. By combining these two methods, the system achieves accurate edge direction handling through the sparse approximation component while avoiding the artifacts that plague pure directional interpolation schemes. The merging of approaches allows the system to benefit from both methods without their respective drawbacks.

Inventive Principle:
Principle #5Merging (Combining)

4Manufacturing precision

If iterative back-projection technique is used to improve image interpolation, then reconstruction accuracy improves when downsampling process is known, but the method fails when downsampling filter is unknown or for camera captured images

Engineering Contradiction:
Improvereconstruction accuracyVSAvoidapplicability to unknown downsampling
Core Design Contradiction:
Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal upsampling method that works regardless of whether the downsampling process is known or unknown. By using sparse approximation to extract high-frequency components directly from the upsampled image without requiring knowledge of the original downsampling filter, the method becomes universally applicable to both synthesized images (where downsampling is known) and camera-captured images (where downsampling is unknown). This universality resolves the contradiction between reconstruction accuracy and adaptability.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP2989606B1Upsampling and signal enhancement
Publication Date: 2020.06.03 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP2989606B1 patent drawingFigure 1~2
  • EP2989606B1 patent drawingFigure 3A~3B
  • EP2989606B1 patent drawingFigure 3C~3D

AI summary

A signal which is to be quality-improved often suffers from the quality degradation in the spatial high frequency region more than compared to the spatial low frequency region. Accordingly a quality improvement is performed efficiently by combining the signal to be quality improved with a high frequency portion extracted from a sparse approximation of the signal to be quality improved.