A Novel Technique for Face Alignment Using Deep Convolutional Neural Networks

Abstract

Face alignment has been neatly done using deep convolutional neural networks. For achieving desired performance, the selecting strategy of local patches for training and their input range is crucial along with configuration of each network. There is double benefit of this. First, the texture background data over the full face is exploited to detect every key point. Second, the training has been given to networks to forecast every key point together; the geometric constraints amid key points are essentially encoded. So, the method can avert local minimum created by uncertainty and data corruption in challenging image samples because of occlusions, broad pose deviations and extreme lightings. Experimental results prove that the approach has high accuracy and is more robust and less complex in nature.