Descriptors Enhancement using Sparse Autoencoder for biometric system based Minor, Major Finger Knuckle pattern
Rabah Hammouche, Abdelouahab Attıa, Samir Akhrouf · 2020
The outer finger knuckle print (FKP) surface has been widely used for personal authentication systems, where the traditional methods for feature extraction used in these systems not be able to achieve interesting results. Thus, we have proposed in this paper to employ a deep learning method named Sparse Auto Encoder (SAE) to enhance the features extracted from several descriptors including Gabor filter, Binarized Statistical Image Features (BSIF), Histograms of Oriented Gradients (HOG), and GIST. The main purpose is to extract features by each descriptor from the input images, then these features are enhanced using the SAE method. Next, these features are reduced by Principal Component Analysis, Linear Discriminant Analysis (PCA+LDA) technique. Then, the Nearest Neighbor (NN) Classifier based on cosine Mahalanobis distance has been applied in the matching stage. The system has been tested on the Contactless Finger Knuckle Images Database (Version 1.0) which are publicly available, the achieved results are good compared to the previous methods given in the state of the art.