Image Processing Apparatus Selective Density Variation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image recognition methods using machine learning require large amounts of training data to account for varying density values of objects in images, but these methods uniformly change density values, which is not suitable for all cases, particularly in medical imaging where certain regions should maintain consistent density values.

Innovation Solution

An information processing apparatus that differentiates regions in an image based on object characteristics and image capturing conditions, selectively changing density values in target regions while keeping non-target regions unchanged, generating new training data that accurately represents real-world imaging conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If uniform density value changes are applied to training data, then the quantity of training data increases, but the manufacturing precision of the training data deteriorates because impossible density values are generated

Engineering Contradiction:

Improvequantity of training dataVSAvoidprecision of training data

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The patent applies local quality by differentiating between target regions and non-target regions in the image. Density value changes are selectively applied only to non-target regions, while target regions maintain their original density characteristics. This resolves the contradiction by generating diverse training data (increasing quantity) without creating impossible density values in critical areas (maintaining precision).

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent segments the image into target regions and non-target regions based on object characteristics and imaging conditions. This segmentation allows differential processing where only non-target regions undergo density value transformation. This approach enables generation of multiple varied training samples while preserving the integrity of target object density values, thus increasing training data quantity without sacrificing precision.

Inventive Principle:

Principle #1Segmentation

2Productivity

If density values are changed to generate new training data, then the productivity of training data generation increases, but the manufacturing precision deteriorates due to unrealistic density values

Engineering Contradiction:

Improveproductivity of training data generationVSAvoidprecision of training data

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

By applying density value changes only to non-target regions rather than uniformly across the entire image, the patent maintains realistic density characteristics in target regions while still generating diverse training data. This localized approach improves productivity through efficient selective processing while preserving precision by avoiding unrealistic density values in critical areas.

Inventive Principle:

Principle #3Local quality

3Ease of manufacture

If uniform density transformation is applied, then the ease of manufacture of training data improves, but the reliability of the training data worsens because it does not accurately reflect real-world conditions

Engineering Contradiction:

Improveease of training data generationVSAvoidreliability of training data

Core Design Contradiction:

Ease of manufactureVSReliability

Solution Approach 1:

The patent maintains ease of manufacture by implementing a systematic automated process that identifies target and non-target regions and applies appropriate transformations. The reliability is improved by ensuring that density value changes are applied selectively and appropriately - only to non-target regions where such variations are realistic, thereby creating training data that accurately reflects real-world imaging conditions.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The segmentation of images into target and non-target regions enables automated processing that is both easy to manufacture (systematic and repeatable) and reliable (accurately reflects real conditions). The segmentation allows the system to automatically determine which regions should undergo density transformation based on object characteristics and imaging parameters, creating realistic training data without manual intervention.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS11961161B2Information processing apparatus, information processing method, and storage medium

Publication Date: 2024.04.16 CANON KK

AI summary

An information processing apparatus according includes an acquisition unit configured to acquire an image, an extraction unit configured to extract part of regions included in the image as a target region, a determination unit configured to determine a variation in density value of a pixel included in the target region such that an absolute value of the variation in density value of the pixel included in the target region is larger than an absolute value of a variation in density value of a pixel included in a region other than the target region, a generation unit configured to generate a training image in which a density value of the pixel included in the target region is changed based on the variation determined by the determination unit, and a training unit configured to train a identifier using the training image.