Kinship verification using Compound Local Binary Pattern and Local Feature Discriminant Analysis
Moumita Mukherjee, Toshanlal Meenpal · 2019
Automatic kinship verification system is designed to verify kinship relations between given pair of face images. Child adopts many characteristic such as similarity in appearance, likes and dislike, behavior, voice from his/her parents due to overlapping of genes. There are various existing algorithms that can verify whether a given pair of face images share kinship relation. This paper proposes a new method based on compound local binary pattern (CLBP) and local feature-based discriminate analysis (LFDA) to improve kinship verification accuracy. A well known texture feature extraction method is local binary pattern (LBP), but LBP performance deteriorates in flat images. To overcome this drawback, extraction of texture features from face images are calculated by compound local binary pattern (CLBP) technique. Extracted features mainly represent facial characteristics but may also contain some noises. Further, local feature-based discriminate analysis (LFDA) is used as a feature selection method to reduce these noises and choose the most relevant facial features. LFDA reduces inter-class similarity and increases intra-class similarity. Proposed method uses KNN classifier with 5-fold cross-validation. Kinship images are collected from KinFaceW-I and KinFaceW-II dataset. Best mean accuracy for proposed method on KinFaceW-I and KinFaceW-II are 82.825 % and 89.36% respectively. Experimental results also outperforms existing methods on these datasets.