Machine Learning Range Estimation for Printed Matter Edge Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional edge detection methods for extracting images of printed matter from captured images are prone to erroneous detection and lack sufficient accuracy, particularly when dealing with faded or dark-colored regions.

Innovation Solution

An image processing apparatus employing a machine learning-based range estimation model to accurately identify the printed matter by setting a wider detection range and using edge detection algorithms to distinguish between the printed matter and adjacent dark-colored regions, thereby enhancing detection precision and reducing erroneous extraction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional edge detection methods are used to extract printed matter from captured images, then the extraction process is simple, but the detection accuracy is low and erroneous detection occurs frequently

Engineering Contradiction:
Improveedge detection accuracyVSAvoiddetection method complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the edge detection process into multiple distinct stages: initial edge detection to identify candidate regions, classification to distinguish printed matter edges from dark-colored region edges, and final edge selection. This segmentation allows each stage to focus on specific tasks, improving overall detection accuracy while managing complexity through modular processing

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary classification step between initial edge detection and final edge selection. This classification process acts as a mediator that filters candidate edges through multiple criteria (color information, positional relationships, continuity) before confirming them as printed matter edges, thereby reducing erroneous detection without requiring complete redesign of the detection system

Inventive Principle:
Principle #24Intermediary (Mediator)

2Area of stationary object

If the detection range is expanded to include more areas, then the printed matter can be fully captured, but the likelihood of detecting dark-colored regions as false edges increases

Engineering Contradiction:
Improvedetection rangeVSAvoidedge detection reliability
Core Design Contradiction:
Area of stationary objectVSReliability

Solution Approach 1:

The patent applies local quality analysis by examining specific properties of edge candidates in different regions. The classification process evaluates color information, positional relationships, and continuity characteristics locally at each candidate edge position, allowing the system to distinguish between valid printed matter edges and false dark-colored region edges even within an expanded detection range

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes multiple parameters simultaneously to improve reliability: color space parameters (using multiple color spaces for classification), spatial parameters (positional relationships between edges and candidate regions), and continuity parameters (whether edges form continuous boundaries). These parameter changes enable reliable distinction between printed matter and dark-colored regions across the expanded detection area

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4425441A1Image processing apparatus, image processing method, program, and recording medium
Publication Date: 2024.09.04 FUJIFILM CORP
  • EP4425441A1 patent drawingFigure 1~2
  • EP4425441A1 patent drawingFigure 3
  • EP4425441A1 patent drawingFigure 4

AI summary

Provided are an image processing apparatus, an image processing method, a program, and a recording medium for more appropriately extracting an image corresponding to a printed matter from a captured image of the printed matter by solving the above-described problem of the related art. An image processing apparatus that extracts an image corresponding to a printed matter from a captured image includes a processor, in which the processor executes an estimation process of estimating a first range including the printed matter in the captured image by applying a range estimation model constructed by machine learning regarding the printed matter to the captured image.