ML Image Tiling With Variable Overlap for Consistent Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing methods using machine learning models face challenges in accurately processing images of varying sizes due to inconsistencies in tile division and overlapping areas, leading to inefficiencies and reduced processing accuracy.

Innovation Solution

A method involving the generation of smaller images with defined overlapping areas and additional padding around borders, followed by processing these images using a machine learning model, and synthesizing the results to maintain consistent pixel counts across tiles, regardless of image size.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If the input image is divided into multiple patches of uniform size, then processing efficiency is improved, but images of varying sizes cannot be processed consistently

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidability to process images of varying sizes
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The input image is divided into multiple patches (tiles) of uniform size to enable efficient parallel processing. Each patch is processed independently by the machine learning model, improving computational efficiency while maintaining the ability to handle images of various original sizes through consistent tiling procedures.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies padding to add pixels to patches with insufficient overlapping areas, and cropping to remove excess pixels from patches with excessive overlapping areas. These parameter adjustments ensure all patches have consistent dimensions and overlapping characteristics, allowing uniform processing regardless of the original image size.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If padding is applied to ensure uniform patch sizes, then processing consistency is improved, but additional computational overhead is introduced

Engineering Contradiction:
Improveprocessing consistencyVSAvoidcomputational overhead
Core Design Contradiction:
Manufacturing precisionVSLoss of energy

Solution Approach 1:

Padding is applied in advance during the patch generation stage, before the machine learning model processes the images. This preliminary action ensures that all patches have the required overlapping areas and uniform dimensions, avoiding the need for complex runtime adjustments and reducing overall computational overhead during the main processing phase.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If overlapping areas are increased to improve boundary processing, then processing accuracy is improved, but the number of pixels to be processed increases

Engineering Contradiction:
Improveprocessing accuracyVSAvoidnumber of pixels
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent applies different overlapping area sizes to different patches based on their position and content requirements. Patches requiring better boundary processing receive larger overlapping areas, while others use standard overlap sizes. This localized approach improves accuracy where needed without uniformly increasing the total pixel count across all patches.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4672138A1Image processing method, image processing apparatus, and storage medium
Publication Date: 2025.12.31 CANON KK
  • EP4672138A1 patent drawingFigure 1A~1D
  • EP4672138A1 patent drawingFigure 2
  • EP4672138A1 patent drawingFigure 3A~3B

AI summary

An image processing method includes generating, by dividing a first image, a plurality of second images each of which is smaller than the first image, and generating a plurality of third images based on the plurality of second images. The plurality of second images includes an image including an overlapping area having a first number of pixels that overlap an adjacent image and an image including an overlapping area having a second number of pixels that overlap an adjacent image, the second number of pixels being different from the first number of pixels. The plurality of third images is generated by using a machine learning model.