Virtual View Image Modeling With Adaptive Spatial Parameters

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing spatial information estimation methods face challenges with excessively large or small spatial information parameter numbers, leading to excessive calculation or insufficient learning accuracy, respectively, in representing objects in virtual viewpoint images.

Innovation Solution

An image processing apparatus that sets parameters in a learning model based on image features and spatial frequencies from multiple viewpoints, adjusting the spatial resolution to match the high-frequency components in captured images, using methods like deep neural networks, 3D Gaussian splatting, or tetrahedron groups to optimize parameter numbers.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the spatial information parameter number is excessively large, then the complexity of spatial information that can be represented increases, but the amount of calculation required for learning becomes enormous

Engineering Contradiction:
Improvespatial information representation capabilityVSAvoidcalculation amount
Core Design Contradiction:
Adaptability or versatilityVSPower

Solution Approach 1:

The patent dynamically adjusts the spatial information parameter number based on the analysis results of captured images. By changing the parameter number according to image characteristics (such as object complexity, scene details), the system achieves appropriate representation capability while avoiding excessive calculation when simple scenes are captured.

Inventive Principle:
Principle #35Parameter changes

2Power

If the spatial information parameter number is too small, then the amount of calculation is reduced, but sufficient learning accuracy cannot be obtained and object representation may be blurred

Engineering Contradiction:
Improvecalculation amountVSAvoidlearning accuracy
Core Design Contradiction:
PowerVSMeasurement precision

Solution Approach 1:

The system analyzes captured images to determine the appropriate spatial information parameter number required for accurate representation. By adjusting parameters based on image complexity analysis, the patent ensures sufficient learning accuracy for complex objects while maintaining efficient calculation for simpler scenes.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The spatial information parameter number is not fixed but dynamically determined based on the characteristics of each captured image. This dynamic adjustment allows the system to optimize between calculation efficiency and learning accuracy for different scenarios.

Inventive Principle:
Principle #15Dynamics

3Productivity

If the spatial information parameter number is not appropriately set, then computational efficiency may be improved, but the representation accuracy of objects in virtual viewpoint images deteriorates

Engineering Contradiction:
Improvelearning efficiencyVSAvoidobject representation accuracy
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent performs preliminary analysis of captured images before the learning process to determine the appropriate spatial information parameter number. This preliminary action enables the system to configure optimal parameters in advance, ensuring both efficient learning and accurate object representation without trial-and-error adjustments.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20260073556A1Image processing apparatus, image processing method, and storage medium
Publication Date: 2026.03.12 CANON KK
  • US20260073556A1 patent drawing
  • US20260073556A1 patent drawing
  • US20260073556A1 patent drawing

AI summary

Parameters in a learning model to represent spatial information are set appropriately. An image processing apparatus 100 according to the present disclosure obtains a plurality of images obtained by image-capturing a three-dimensional space containing an object from multiple directions, analyzes the images to obtain an image feature related to each of the images, obtains position information indicating a position of the object, and sets parameters in a learning model to estimate spatial information related to a training region contained in the three-dimensional space, based on the position information and the image feature.