CNN Parameter Training via Statistical Atlas for Fingerprint Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data augmentation techniques for convolutional neural networks (CNNs) are inadequate for effectively training models to classify digital fingerprints, as they fail to account for the specific nonlinear distortions and variability inherent in fingerprint data, leading to poor performance.
Innovation Solution
A method that generates alternate versions of training data by applying geometric and intensity transformations based on a statistical distribution of observed transformations in the training database, using a statistical atlas to model and apply transformations that respect the natural variability of fingerprint data, thereby creating a more realistic and varied training dataset.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional data augmentation techniques (random intensity or geometric modifications) are applied to training data, then the volume of training data is artificially increased, but the classification performance remains poor because the specific nonlinear distortions and variability of fingerprint data are not accounted for
Solution Approach 1:
The patent applies specific geometric transformations (rotation, scaling, translation, skewing) and intensity transformations (contrast adjustment, brightness adjustment, noise addition) to generate augmented fingerprint training data. These parameter changes create realistic variations in the training dataset that account for the natural variability and distortions present in actual fingerprint data, thereby improving classification performance while increasing data volume
Solution Approach 2:
The patent creates multiple copies of existing fingerprint training data by applying transformation operations. Each original fingerprint image is copied and transformed according to predefined geometric and intensity parameters, generating synthetic variations that expand the training dataset without requiring additional real fingerprint samples
2Adaptability or versatility
If the CNN is trained on a larger volume of training data using conventional augmentation, then overtraining is avoided and generalization is improved, but the specific variability and nonlinear distortions of fingerprint data are not properly represented
Solution Approach 1:
The patent defines specific ranges and distributions for transformation parameters (rotation angles, scaling factors, contrast adjustments) that reflect the actual variability observed in real fingerprint data. By carefully controlling these parameter changes, the augmented data maintains realistic fingerprint characteristics while providing sufficient diversity for good generalization
Solution Approach 2:
The patent applies different types of transformations to different regions or aspects of the fingerprint data. Geometric transformations address spatial variability and distortion, while intensity transformations address lighting and contrast variations. This localized application of quality enhancements ensures comprehensive representation of fingerprint variability
Data Source
AI summary
A parameter training method for a convolutional neural network (CNN) for classifying image type data representative of a biometric trait. The method includes the implementation, by a data processor of a server, the steps of (a) for at least one data item from an already classified training database, generation of several alternate versions of this data each by application of at least one transformation chosen from a set of reference transformations satisfying a statistical distribution of transformations observed in the training database and (b) training the parameters of the CNN, from the already classified training database augmented with said alternate versions of the data.