ML Image Tiling With Variable Overlap for Boundary Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate 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 in processing performance.

Innovation Solution

A method involving padding and mirroring to ensure consistent tile division and overlapping pixels, followed by sequential processing and weighted averaging or concatenation to generate high-quality output images, regardless of input 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 processing accuracy deteriorates due to loss of contextual information at boundaries

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidprocessing accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The input image is divided into multiple patches of uniform size for parallel processing, improving efficiency. The segmentation is performed systematically by calculating patch coordinates and sizes based on image dimensions and a specified patch count, enabling efficient distributed processing while maintaining manageable data sizes.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Padding pixels are added to the input image before dividing it into patches. This preliminary action ensures that patches near the image boundaries have sufficient contextual information, preventing loss of edge details and improving processing accuracy without compromising the efficiency gains from uniform patch division.

Inventive Principle:

Principle #10Preliminary action

2Stability of the object's composition

If padding is applied to unify patch sizes, then processing consistency is improved, but computational overhead increases

Engineering Contradiction:

Improveprocessing consistencyVSAvoidcomputational overhead

Core Design Contradiction:

Stability of the object's compositionVSUse of energy by stationary object

Solution Approach 1:

Padding is applied selectively and locally only to regions where it is needed - specifically at the boundaries of the input image and around patches that would otherwise be truncated. The padding operation modifies only the necessary pixels rather than the entire image, minimizing computational overhead while ensuring processing consistency across all patches.

Inventive Principle:

Principle #3Local quality

3Loss of information

If overlapping areas between adjacent patches are increased, then information completeness is improved, but data redundancy increases

Engineering Contradiction:

Improveinformation completenessVSAvoiddata redundancy

Core Design Contradiction:

Loss of informationVSQuantity of substance

Solution Approach 1:

The patent applies a moderate degree of overlapping between adjacent patches - not minimal (which would lose boundary information) and not maximal (which would create excessive redundancy). The overlap is carefully controlled to include sufficient contextual information from adjacent regions while keeping the additional data volume manageable, achieving an optimal balance between information completeness and redundancy.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20260004470A1Image processing method, image processing apparatus, and storage medium

Publication Date: 2026.01.01 CANON KK

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.