Local Binary Pattern in the Frequency Domain: Performance Comparison with Discrete Cosine Transform and Haar Wavelet Transform

This study presents a new method that aims to improve iris recognition performance by amplifying high-frequency components in the frequency domain, considering that iris images naturally contain high-frequency details. The Haar Wavelet Transform (HWT) and Discrete Cosine Transform (DCT) are used to enhance these components and an inverse transformation is applied to obtain iris images with more details. As input, the brightness values of the 8 neighboring pixels around each central pixel are used. These values are transformed into the frequency domain, the high-frequency band is amplified, and the data are reconstructed. Feature vectors are then generated using the Local Binary Pattern (LBP) algorithm, which is fed with the enhanced images. These feature vectors are formed using a combination of local histograms rather than a global LBP histogram, which are normalized to ensure consistency. The generated feature vectors are divided into a 70% training set and a 30% test set and are tested using K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Random Forest (RF) algorithms. The proposed method provides a significant performance improvement in terms of accuracy compared to traditional approaches. While both HWT and DCT yield similar results, it has been observed that HWT is much faster. In this study, a comparison is made in terms of both speed and accuracy. Two different public iris datasets, MMU1 and MMU2, are used. This work not only introduces an innovative approach to iris recognition, but also makes a significant contribution to the manipulation of pixel brightness values in the frequency domain, with the findings being expected to guide future research.