Face recognition from image patches using an ensemble of CNN-local mesh pattern networks

Face recognition is one of the superior methods among the biometric identification system. Demand for face recognition is increased recently because of the availability of data from security and surveillance cameras. This work proposes a face recognition method using an ensemble of convolutional neural network trained on original and the textural feature maps from image patches. The original face image divided into smaller patches and a textural descriptor is applied on them. The textural feature map image of all patches and the original patches are input to two pre-trained ResNet50 networks. Input images are used to fine-tune the networks. The predictions of the two networks are combined by majority voting ensemble technique to get the final labels. The proposed system achieves an accuracy of 98.05%. The system performance is also evaluated using different measures like precision, recall, and F-measure.