Low-Dimensional Shape Descriptors via Diffraction Features

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing shape representation methods in engineering design, such as direct parameterizations and deformation methods, face challenges in applying machine learning techniques due to varying parameter meanings across different datasets, leading to inefficiencies and information loss, especially when dealing with high-dimensional geometry data.

Innovation Solution

A computer-implemented method generates a set of unified, low-dimensional diffraction feature shape descriptors by calculating characteristic interference patterns from spherical waves emitted by shape points, allowing for efficient representation and processing of two or three-dimensional shapes without losing local or global geometric information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If coordinates of the entire shape geometry are used as input parameters for machine learning, then complete geometric information is preserved, but the dimensionality of the input parameter space becomes very high requiring huge training datasets

Engineering Contradiction:
Improvegeometric informationVSAvoidinput parameter space dimensionality
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent extracts essential geometric information from complete shape coordinates by computing diffraction features at selected feature locations. Instead of using all coordinate data, it extracts a subset of characteristic features that capture the essential geometric properties while discarding redundant information, thereby reducing dimensionality while preserving key geometric characteristics.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces diffraction features as an intermediary representation between raw shape coordinates and machine learning inputs. These diffraction features serve as a mediator that transforms high-dimensional coordinate data into a lower-dimensional feature space that retains essential geometric information while being suitable for machine learning processing.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If shape parameters are used as input for machine learning techniques, then processing efficiency is improved, but the meaning of parameters changes across different datasets making direct application impossible

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidparameter meaning consistency
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal shape representation method using diffraction features that can be applied across different shape datasets regardless of their original parameterizations. The diffraction feature computation is dataset-agnostic and produces consistent features for any input shape, enabling machine learning models to process diverse shape data uniformly without needing dataset-specific adaptations.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent transforms shape representations from dataset-specific parameters to a unified diffraction feature space. By changing the parameter representation from original shape parameters to diffraction features computed at standardized feature locations, the method achieves parameter meaning consistency across different datasets while maintaining processing efficiency.

Inventive Principle:
Principle #35Parameter changes

3Device complexity

If global shape descriptors are used to reduce dimensionality, then computational efficiency is improved, but local geometric information is lost due to voxelization or signature derivation

Engineering Contradiction:
Improvefeature vector dimensionalityVSAvoidlocal geometric information
Core Design Contradiction:
Device complexityVSLoss of information

Solution Approach 1:

The patent applies local quality by computing diffraction features at multiple distributed feature locations around the shape rather than using a single global representation. Each feature location captures local geometric characteristics, and the collection of all feature locations provides both local and global shape information. This approach preserves local geometric details while maintaining computational efficiency through dimensionality reduction.

Inventive Principle:
Principle #3Local quality

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This method provides a cost-effective, low-dimensional representation of shapes that enables efficient computational processing and data management, maintaining complete information and being sensitive to local changes, thus facilitating effective storage, classification, and optimization tasks in engineering applications.

Implementation Method 1

calculating diffraction features for a set of feature locations distributed around the shape

Methodology Applied
Scientific EffectDiffraction: Diffraction

Implementation Method 2

calculating characteristic interference patterns from spherical waves emitted by shape points

Methodology Applied
Scientific EffectInterference: Interference

Data Source

PatentEP3637319B1Method for generating shape descriptors for two- or three-dimensional geometric shapes
Publication Date: 2022.05.25 HONDA RES INST EUROPE
  • EP3637319B1 patent drawingFigure 1
  • EP3637319B1 patent drawingFigure 2
  • EP3637319B1 patent drawing

AI summary

In the invention for generating a set of shape descriptors for a set of two or three dimensional geometric shapes in order to arrive at an unified efficient low-dimensional representation of the complete set of shapes to enable memory and disk efficient storage, indexing, referencing, and making the complete set available for further processing, at first a set of N feature locations having a distance from the shapes is read. Further, a set of M wave numbers is read and a parameter controlling degree of locality of the features. Then, for each shape s in the set of shapes {Ss, s = 1, ..., Ns} and for each of the N feature locations and M wave numbers a feature descriptor is calculated according to fsmm=‖R→n‖aγe−ikmRnC∫shape sd3s→eikm‖s→−R→n‖2‖s→−R→n‖αγ where the integral is summing all contributions from each point of shape s. The calculated feature descriptors are then assigned to elements of an M.N dimensional vector as the shape descriptor for shape s F→s=fsn=1,m=1,fsn=1,m=2,...,fsn=N,m=MT and the complete set of shape descriptors {Fs, s = 1, ... , Ns} of the set of shapes is output.