Feature Point Descriptor Using Ring Sectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image analysis methods for scene analysis rely heavily on proprietary feature point detectors and descriptors, necessitating licensing and lacking alternative scale-invariant solutions.

Innovation Solution

A computer-implemented method that describes the surrounding of a feature point using a feature vector by dividing the image into non-overlapping rings with sectors, allowing for multiscale descriptor representation and improved robustness to noise and rotation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If proprietary feature point detectors and descriptors (e.g., SIFT, SURF) are used, then accurate feature point detection and description is achieved, but licensing requirements and increased cost are imposed

Engineering Contradiction:
Improvefeature point detection accuracyVSAvoidlicensing requirement
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent implements alternative feature point detection and description methods that replicate the functionality of proprietary algorithms like SIFT and SURF without requiring their licenses. The invention uses publicly available algorithms and approaches to create comparable feature vectors, thereby achieving accurate feature point matching while avoiding licensing costs and restrictions

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent employs computationally efficient feature detection and description techniques that can be executed without expensive proprietary software licenses. By using alternative algorithms that are freely implementable, the system achieves comparable performance to proprietary solutions while eliminating licensing overhead and costs

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

2Reliability

If traditional feature descriptors are used, then feature point description is achieved, but scale invariance and noise resistance are insufficient

Engineering Contradiction:
Improvenoise resistanceVSAvoiddescriptor structure
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent divides the feature point description into multiple segments or components, where each segment captures specific aspects of the local image structure. This segmentation approach enhances robustness to noise and scale variations by distributing information across multiple descriptor elements, allowing the system to maintain reliability even when individual segments are affected by noise

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extends feature descriptors into additional dimensions to capture scale-invariant information. By incorporating multi-scale analysis and adding dimensional components that represent features at different scales, the descriptor achieves scale invariance while maintaining a structured organization that manages complexity through hierarchical arrangement

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Adaptability or versatility

If existing feature point methods are used, then image comparison is achieved, but alternative scale-invariant solutions are lacking

Engineering Contradiction:
Improvescale invarianceVSAvoidmethod alternatives
Core Design Contradiction:
Adaptability or versatilityVSLoss of information

Solution Approach 1:

The patent develops feature detection and description methods that are universally applicable across different image scales and conditions. The implemented algorithms provide multi-functional capability by detecting and describing features that remain consistent across scale transformations, thereby achieving scale invariance and providing versatile solutions that work across diverse image analysis scenarios

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

Data Source

PatentEP3113077B1Method for image feature point description
Publication Date: 2018.08.29 LINGARO
  • EP3113077B1 patent drawingFigure 1A~1B
  • EP3113077B1 patent drawingFigure 2A
  • EP3113077B1 patent drawingFigure 2B

AI summary

A computer-implemented method for describing a surrounding of a feature point (F) in an image by a feature vector (D), the method comprising: assigning (101) the pixels of the image located around the feature point (F) to at least two segments, each segment having a form of a non-overlapping ring centered at the feature point (F), wherein the ring of the first segment consists of pixels located not farther than a first radius (R1) from the feature point (F); and wherein the ring of the second segment consists of pixels located not farther than a second radius (R2) from the feature point (F) and that do not belong to the first ring; assigning (102) the pixels of each segment to angular sectors (S1, S2, ...), the sectors comprising pixels arranged within a particular angular range from the feature point (F), the angular range being equal for each of the sectors (S1, S2,...), wherein the sectors are adjacent to each other or overlap each other by the same overlapping range, wherein each of the pixels belongs to no more than two sectors; within each sector of each segment, calculating (103) at least one descriptor measure as a measure function of values of pixels belonging to that sector; and providing (104) the feature vector as a set of the calculated descriptor measures for all sectors of all segments.