Image Upscaling Using Regression Functions and Local Ternary Patterns

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for upsampling low-resolution images to high-resolution images suffer from artifacts like blurry edges, exaggerated noise, halos, ringing, and aliasing, and require significant computational resources, making them unsuitable for real-time applications.

Innovation Solution

The method uses regression functions derived from pairs of high-resolution and low-resolution patches within clusters based on local n-ary pattern signatures to estimate high-resolution patches, employing off-line training and real-time upscaling, which generates high-resolution images without losing fine textural details.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If parametric or non-parametric upscaling methods are used to generate high-resolution images from low-resolution images, then image resolution is improved, but computational resources required increase significantly

Engineering Contradiction:
Improveimage resolutionVSAvoidprocessing speed
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The patent pre-computes and stores correspondence relationships between low-resolution and high-resolution image patches in an offline training phase. During real-time upscaling, the system only needs to perform simple lookup and composition operations using pre-built regression functions, dramatically reducing online computational requirements while maintaining high resolution output

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent divides the image upscaling problem into independent patch-level regression tasks. By processing images as collections of small patches rather than entire images, the computational complexity is reduced from O(N²) to O(N) where N is the number of pixels, enabling real-time processing while achieving high resolution reconstruction

Inventive Principle:
Principle #1Segmentation

2Area of stationary object

If conventional upscaling methods are used to increase image resolution, then image size is improved, but image quality deteriorates due to artifacts

Engineering Contradiction:
Improveimage sizeVSAvoidimage quality
Core Design Contradiction:
Area of stationary objectVSManufacturing precision

Solution Approach 1:

The patent learns correspondence relationships by copying patterns from high-resolution reference images to low-resolution input images. The regression functions are trained to copy fine textural details and edge structures from HR patches, enabling the synthesis of realistic high-resolution textures that are not present in the original LR image

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms the upscaling problem from direct pixel interpolation to a parameter-based regression problem. By representing image patches in terms of learned parameters (regression coefficients) rather than direct pixel values, the system can generate high-resolution outputs with realistic textures while avoiding the artifacts of conventional interpolation methods

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3120322B1Method for processing input low-resolution (LR) image to output high-resolution (HR) image
Publication Date: 2019.04.24 MITSUBISHI ELECTRIC CORP
  • EP3120322B1 patent drawingFigure 1
  • EP3120322B1 patent drawingFigure 2
  • EP3120322B1 patent drawingFigure 3

AI summary

A method generates a high-resolution (HR) image from a low-resolution (LR) image using regression functions. During a training stage, training HR images are downsampled to LR images. A signature is determined for each LR-HR patch pair based on a local ternary pattern (LTP). The signature is a low dimensional descriptor used as an abstraction of the patch pair features. Then, patch pairs with the same signature are clustered, and a regression function which maps the LR patches to the HR patches is determined. In some cases patch pairs of similar signatures can be combined for learning and a single regression function determined, thus decreasing the number of required regression functions. During actual upscaling, LR patches of an input image are similarly processed to obtain the signatures and from the regression functions. The LR patches can then be upscaled using the training regression functions.