A distributed efficient face image screening and retrieval with affine transformations, disguised, and varying facial expressions

The detection of wanted or suspicious individuals, encountered generally in airports or transportation stations, via the recognition of their faces for security purposes, is an important area for preventing terrorism and smuggling. This problem remains still in its infancy due to the many challenges involved. In particular, face recognition becomes more complicated and challenging when a given face to recognize is rotated off the plane, or is having changes in the facial expressions, or is being disguised, as it is often the case in the real world. In this paper we consider the problem of, given a time-varying face image template or only a sub-part of it, whether the template object image exists in a given image database of images stored in a given nominal configuration. We demonstrate here the main advantages of using sub-image patches and patches pre-processing instead of the whole face image are three-fold namely (a) the ability to deal with the situation of face disguise where only a small portion of the face can be recognized, (b) smaller face sub-image patches are less tuned to being significantly affected by affine transformations such as image translation/scaling/off-the-plane-rotation (c) the ability to perform image searching in an efficient way; since correlation methods have their short coming of requiring extremely large computation costs caused by exhaustive database searches and image similarity computations, when the image collections are large. In this work we demonstrate how by using sub-image patches leads itself to a distributed architecture for face recognition and how by combining a technique of distributed computing with a an image template pre-processing method the performance of image searching and matching is greatly improved by a several order of magnitude.