Adaptive Multi-Subband Wavelet–Phase Fusion for Automated Down Syndrome Detection from Facial Images

Abstract

Early identification of Down Syndrome (DS) plays a vital role in securing timely medical intervention as well as in planning for long, term care. To address this issue, the current work offers an automated framework that performs facial image analysis using adaptive multi subband feature fusion in the frequency domain. The proposed method works by first decomposing images with the Haar wavelet transform to capture details at various resolutions, then applying Local Phase Quantization (LPQ) to extract texture features that remain reliable even when images are blurry or lighting is poor. Then an adaptive weighting scheme is utilized to integrate the features, thereby amplifying the system's capability to correctly identify the relevant patterns prior to the classification with the Support Vector Machine (SVM). The performance of this solution was compared to the traditional descriptors such as Local Binary Patterns (LBP), Histogram of Oriented Gradients (HOG), and geometric landmarks. The data demonstrate that the adaptive fusion method attains an accuracy of 94. 6%, a sensitivity of 95. 1%, and a specificity of 94. 0%, thus surpassing the other methods. These results imply that multi resolution wavelet, phase modeling is a strong, easily understandable, and efficient tool for the automated early screening of Down Syndrome.