A Novel Approach for Recognition of Face by Using Squeezenet Pre-Trained Network

A face recognition in current generation plays a vital role. Day by day new techniques are emerging for face recognition. Now a days deep learning based face recognition techniques have become very popular due to the less recognition time and fast computation by using the sequence of layers and drop out is employed for training the deep neural network for some layers for reducing the over fitting of data in neural networks. Leveraging ready-to-use pre-trained deep neural networks for specific applications, some of the layers are configured with transfer learning to perform face recognition tasks, significantly simpler and more accurate than previous techniques too. The squeezenet pre-trained convolutional neural network is used here by replacing two final layers with new layers adjusted to new data set. The Experimentation is performed on data sets such as ORL and Extended YALE B. Data is partitioned as training and validation datasets. 80% of image as training, 20% of image data is used as validating data. 97% of accuracy is achieved on training data and 92% of accuracy is obtained on validation data.