Kinship Verification Using Multiscale Retinex Preprocessing and Integrated 2DSWT-CNN Features

Kinship verification from facial images presents a challenging yet intriguing problem within the fields of pattern recognition and computer vision. In this study, we introduce significant advancements by applying a preprocessing technique known as Multiscale Retinex with Color Restoration (MSRCR) to enhance the quality and contrast of images. Our methodology uniquely combines the strengths of both deep and shallow feature descriptors by integrating them at the score level through the use of Logistic Regression (LR). Specifically, we utilize a novel descriptor, Histograms of a Two-Dimensional Stationary Wavelet Transform (Hist-2D-SWT), to capture non-deep features, while employing the CNN model for deep feature extraction. The efficacy of our approach is thoroughly evaluated through extensive experiments conducted on three challenging kinship datasets: Cornell KinFace, UB KinFace, and TS KinFace.