Fingerprint Image Reconstruction from Sensor Fragments

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional fingerprint recognition algorithms are inefficient when dealing with partial fingerprint images from small sensors due to image deformations and reduced quality, leading to inaccurate and unreliable fingerprint recognition in devices like smartphones and smart cards.

Innovation Solution

A method for reconstructing a complete fingerprint image from multiple overlapping image parts using local interest points, descriptor vectors, and an iterative assembly process to evaluate and merge image parts based on association scores, reducing distortions and storage requirements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Area of stationary object

If multiple overlapping image parts are acquired to cover the entire fingerprint on a small sensor, then the complete fingerprint image can be obtained, but image distortions and deformations occur during acquisition and reconstruction

Engineering Contradiction:
Improveacquisition areaVSAvoidimage quality
Core Design Contradiction:
Area of stationary objectVSManufacturing precision

Solution Approach 1:

The fingerprint image is divided into multiple overlapping fragments captured by a small sensor. Each fragment contains local interest points and descriptor vectors that are processed independently, then assembled together to reconstruct the complete fingerprint image, resolving the contradiction between limited sensor area and complete image coverage

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Descriptor vectors serve as intermediaries to match and align overlapping image fragments. By comparing descriptor vectors between adjacent fragments, the system can accurately stitch images together while compensating for distortions, maintaining image quality despite the segmentation approach

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If classic recognition algorithms are used to compare complete fingerprint images, then recognition accuracy is high, but they are ineffective when dealing with image fragments from small sensors

Engineering Contradiction:
Improverecognition accuracyVSAvoidalgorithm effectiveness
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

Instead of applying global recognition algorithms to complete images, the system extracts and compares local interest points and their descriptor vectors from each image fragment. This local approach adapts the recognition process to work effectively with fragmented images while maintaining accuracy by focusing on distinctive local features

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system transforms the recognition problem from comparing complete 2D fingerprint images to comparing sets of local interest points and their descriptor vectors. This dimensional transformation allows classic recognition algorithms to be applied to fragment data, bridging the gap between fragment acquisition and complete image recognition

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

3Measurement precision

If iterative assembly process with association scores is used to reconstruct fingerprint image, then recognition accuracy improves, but computational complexity increases

Engineering Contradiction:
Improvereconstruction precisionVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system pre-calculates and stores descriptor vectors for all local interest points in each image fragment before the assembly process. This preliminary action allows for efficient comparison and matching during iterative assembly, reducing real-time computational complexity while maintaining high reconstruction precision through association scores

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3621026B1Method for reconstructing an imprint image from image portions
Publication Date: 2022.07.20 IDEMIA IDENTITY & SECURITY FRANCE SAS
  • EP3621026B1 patent drawingFigure 1~2
  • EP3621026B1 patent drawingFigure 3
  • EP3621026B1 patent drawingFigure 4

AI summary

Method for reconstructing a fingerprint image (12), representative of a papillary print, from a set of image parts (1, 2, 10) acquired using at least one sensor, comprising the steps of: - extracting from each image part a set of local points of interest (step E1) and, for each local point of interest, calculating a descriptor vector that characterizes said local point of interest; - for each pair of two image parts, evaluating an association score of the local points of interest representative of a probability that the two image parts are contiguous on the fingerprint image; - assembling the image parts of a best pair, whose association score is the highest, to form an assembled fragment (step E4);- repeat the above steps, each time replacing the two image parts of the best pair with the assembled fragment in the set of image parts, until all the association scores of the remaining pairs are less than or equal to a predetermined threshold, thus producing an assembly map of the image parts; - merge the image parts according to the assembly map obtained during the assembly steps to reproduce the fingerprint image (step E5).