Robust Face Alignment Using Convolutional Neural Networks

Face recognition in real-world images mostly relies on three successive steps: face detection, alignment and identification. The second step of face alignment is crucial as the bounding boxes produced by robust face detection algorithms are still too imprecise for most face recognition techniques, i.e. they show slight variations in position, orientation and scale. We present a novel technique based on a specific neural architecture which, without localizing any facial feature points, precisely aligns face images extracted from bounding boxes coming from a face detector. The neural network processes face images cropped using misaligned bounding boxes and is trained to simultaneously produce several geometric parameters characterizing the global misalignment. After having been trained, the neural network is able to robustly and precisely correct translations of up to ±13% of the bounding box width, in-plane rotations of up to ±30◦ and variations in scale from 90% to 110%. Experimental results show that 94% of the face images of the BioID database and 80% of the images of a complex test set extracted from the internet are aligned with an error of less than 10% of the face bounding