Performance Analysis of Multiple Wavelets for Robust Half-Iris Verification with SVM

Our experiments demonstrate that fusing features extracted from multiple wavelets leads to improved verification accuracy for segmented half-iris templates compared to single-feature methods. Support Vector Machines (SVMs) were employed for classification.We evaluated the performance of nineteen different wavelets across various decomposition levels of the Discrete Wavelet Transform (DWT). When we consider only one wavelet feature, the coiflet4 wavelet at the second-level horizontal decomposition achieved the highest accuracy of 92%. Conversely, the biorthogonal 1.3 wavelet at the fourth-level horizontal decomposition yielded the lowest accuracy at 65%. These findings highlight the importance of selecting appropriate wavelets for optimal performance.When we combined multiple features, the verification accuracy increases upto of 98% for CASIA and UBIRIS databases. Hence, SVM classification with combined wavelet features proves effective for iris-based person verification.