Deep Multi-Patch Aggregation Network for Kinship Recognition

Recognizing genetic identities through face images can be used as an implementation of face recognition systems. However, recent developments in face detection have also shown that there is still more to learn from the use of more data and new innovations. Mostly, facial recognition is a good source domain from that we can transfer information to obtain better outcomes for kinship recognition and generate a family tree. In this paper, we address the identification of kinship through face images while considering different patches by using feature extraction methods, immediately labeling pairs in face images to identify the relationships. The proposed method is applied on the dataset Families In the Wild (FIW). Eventually, a group of classification problems with low-level image functions are introduced and estimated. Several state of the art architectures are used considering the multi-patch technique for features extraction. While identifying another very distinct inherited facial features, we successfully demonstrate classification performance of 85.11% on a test set using Multi Patch Deep Convolutional Neural Network (MPD-CNN) architecture of image combinations.