Neural Network Defocus Estimation for Obstruction Handling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing technologies face challenges in accurately focusing on a subject, especially when obstructions like arms or hands cover the face, and in low-light or high-contrast environments, leading to errors in focus detection due to variations in distance values.

Innovation Solution

An image processing apparatus utilizing a neural network to estimate defocus amounts from defocus maps, integrating features generated from images, defocus maps, and subject region maps to improve focus adjustment accuracy, while reducing the influence of obstructions and focus detection errors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If statistical values of distance values are used to determine obstruction regions, then focus adjustment can be performed, but measurement precision deteriorates due to variations in distance values in low-light or high-contrast environments

Engineering Contradiction:
Improvefocus adjustment reliabilityVSAvoiddistance value measurement precision
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent introduces a neural network as an intermediary component that processes defocus map data to estimate subject defocus amounts. Instead of directly using statistical values of distance values (which are prone to errors in low-light/high-contrast conditions), the neural network acts as a mediator that transforms the defocus map into reliable defocus amount estimates, filtering out the influence of obstructions and environmental variations.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the traditional mechanical/statistical approach (using statistical values of distance values) with a neural network-based system. This substitution allows the system to process defocus map information through learned patterns rather than relying on direct statistical calculations, thereby improving measurement precision in challenging lighting conditions.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If defocus amounts are estimated using traditional methods, then processing speed is maintained, but measurement precision deteriorates in low-light or high-contrast environments

Engineering Contradiction:
Improvedefocus amount estimation precisionVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent performs preliminary action by pre-processing the defocus map data through neural network training before actual focus adjustment is needed. The neural network learns patterns from training data and can quickly estimate defocus amounts during inference without requiring complex real-time calculations, thus achieving high precision while maintaining processing speed.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If obstructions are present in the image, then focus detection can still be performed, but measurement precision deteriorates due to errors in distance values

Engineering Contradiction:
Improvefocus detection adaptability to obstructionsVSAvoiddistance value precision
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent extracts and eliminates the influence of obstructions from the focus detection process. By using neural networks to process defocus map data, the system can identify and exclude regions affected by obstructions (such as arms or hands blocking the face) from the distance value calculations, thereby maintaining measurement precision even when obstructions are present in the image.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4465241A1Image processing apparatus, image processing method, and computer program
Publication Date: 2024.11.20 CANON KK
  • EP4465241A1 patent drawingFigure 1
  • EP4465241A1 patent drawingFigure 2
  • EP4465241A1 patent drawingFigure 3

AI summary

An image processing apparatus (2, 13) is provided. The apparatus acquires (S301, S302, S303, S901) input data including a captured image and/or information relating to the captured image. The apparatus acquires (S305, S903) a feature of the input data by performing processing on the input data using a neural network. The apparatus generates (S306, S904) an integrated feature by integrating the feature and at least some of the input data. The apparatus generates (S307, S905) an estimation result of at least one of a defocus range and a depth range for a subject within the captured image, by performing processing on the integrated feature.