ENHANCING IMAGE FINGERPRINT THROUGH WAVELET TRANSFORM METHOD: AN EVALUATION WITH CONVOLUTIONAL NEURAL NETWORKS (CNN)

The problem of low image quality is even worse with multiple fingerprints
because the recognition process depends heavily on the quality of the fingerprint.
In general, fingerprints can be divided into seven types. Fingerprint image
enhancement is an essential preprocessing technique in fingerprint-based personal
authentication systems. This study aims to design a fingerprint image
enhancement system using the CNN model to classify fingerprints into seven
classes and then analyze the effect of fingerprint image enhancement on CNN
performance and classification. Based on the Two Dimension Discrete Wavelet
Transform (2D-DWT), Singular Value Decomposition (SVD) and Otsu Threshold,
adjustments have been applied to the image processing system, significantly
reducing noise and improving image quality because it can control the value
contained in each image pixel. The enhanced image is used as a dataset in the
classification process, namely the Convolutional Neural Network (CNN). This
study's results indicate an increase in the level of accuracy produced by the CNN
module from the original image with the enhanced image. The result of
comparing the accuracy value between the original image and the enhanced image,
the enhanced image has a higher accuracy value compared to the original image.
In the FVC2002 fingerprint dataset, the accuracy value of the original image is
only 65% to 87% after the enhancement process. In the FVC2004 dataset, the
accuracy value of the original image is only 58% to 81% after enhancement. And
in the NIST fingerprint data set, the original accuracy was 42% to 88% after
refinement, so the enhanced image can affect the accuracy performance.
Indeks terms: fingerprint, image enhancement, two dimensions discrete wavelet
transform, singular value decomposition, convolutional neural network.