Neural Network Image Up-Conversion via Block Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image up-conversion techniques face a tradeoff between speed and quality, particularly in real-time applications where image quality may suffer to achieve faster processing.

Innovation Solution

A computer system that uses a trained neural network to convert images from a first resolution to a second resolution by dividing the image into blocks, adding context data, and applying separable block transforms to generate output channels that construct the image in the higher resolution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If traditional up-conversion techniques are used to achieve fast processing, then speed is improved, but image quality deteriorates

Engineering Contradiction:
Improveprocessing speedVSAvoidimage quality
Core Design Contradiction:
SpeedVSManufacturing precision

Solution Approach 1:

The input image is divided into multiple blocks, and each block is processed independently through the neural network. This segmentation allows parallel processing of multiple blocks simultaneously, maintaining high processing speed while applying quality-enhancing transformations to each block individually through the neural network model.

Inventive Principle:
Principle #1Segmentation

2Loss of time

If real-time processing is implemented, then processing time is reduced, but computational complexity increases

Engineering Contradiction:
Improveprocessing timeVSAvoidcomputational complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

Dividing the image into blocks enables parallel processing architecture where multiple blocks are transformed simultaneously. This reduces overall processing time while distributing computational complexity across multiple independent operations that can be executed in parallel, making real-time processing feasible.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The neural network model is trained in advance on large datasets to learn optimal transformation patterns. During real-time inference, the pre-trained model applies learned transformations directly to image blocks without requiring complex real-time computations, significantly reducing processing time while maintaining quality.

Inventive Principle:
Principle #10Preliminary action

3Manufacturing precision

If high-quality up-conversion is achieved through advanced techniques, then image quality is improved, but processing speed decreases

Engineering Contradiction:
Improveimage qualityVSAvoidprocessing speed
Core Design Contradiction:
Manufacturing precisionVSSpeed

Solution Approach 1:

Processing images in blocks allows the neural network to focus computational resources on smaller, manageable units. This enables application of complex quality-enhancing transformations to each block individually while maintaining overall processing efficiency through parallel execution of multiple blocks.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The neural network applies transformations selectively to enhance image quality where needed, rather than uniformly processing all pixels. The model learns to apply appropriate levels of enhancement based on local image characteristics, achieving high quality without unnecessary computational overhead.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4128043B1Systems and methods for machine learned image conversion
Publication Date: 2025.01.29 NINTENDO CO LTD
  • EP4128043B1 patent drawingFigure 1
  • EP4128043B1 patent drawingFigure 2
  • EP4128043B1 patent drawingFigure 3

AI summary

A computer system is provided for converting images through use of a trained neural network. A source image is divided into blocks and context data is added to each pixel block. The context blocks are split into channels and each channel from the same context block is added to the same activation matrix. The action matrix is then executed against a trained neural network to produce a changed activation matrix. The changed activation matrix is then used to generate a converted image.