DWT ‐ DCSIFT : A Novel Feature Representation for Facial Kinship Verification

Facial kinship verification is a significant area of research that aims to determine whether two people are biologically related based on their facial similarity. This study proposes a new feature representation for facial kinship verification, named Discrete Wavelet Transform‐Dense Color Scale Invariant Feature Transform (DWT‐DCSIFT), which combines the frequency, color, and texture descriptions of the face image into a single representation. The frequency information is provided by the 2D‐DWT, allowing spatial and detail features to be captured at different resolutions. The color and texture information is provided by the DCSIFT descriptor, which uses a dense color histogram to exploit the color information of the face image, and a dense SIFT algorithm to exploit the facial texture features. The combined DWT‐DCSIFT feature vector incorporates the spectral information representing the shape features, the color information, and the texture details of the face image, leading to a richer representation for kinship verification. The Fisher score algorithm selects the top‐ranked features, which are then subjected to dimensionality reduction using principal component analysis. The resulting data is then fed into the support vector machines classifier for the final decision. The proposed kinship verification approach produced competitive results and demonstrated enhanced performance over state‐of‐the‐art techniques. In particular, the proposed model achieved high accuracy in discriminating between kin and non‐kin relationships, reaching 81.60% for the KinFaceW‐I dataset and an impressive 91.05% for the KinFaceW‐II dataset.