F acial recognition is a subject of great importance and a lot of literature is already dedicated to it. Recognition of non-rigid surfaces such as faces is a difficult task to tackle both at an inter-personal and intra-personal level, mostly due to variation in one's facial structure over time. The problem is further complicated by the addition of non-rigid elements, notably the introduction of a wide range of facial expressions, which are controlled by many minuscule muscles and can vary enormously by the combination of these muscles' state. With the growing interest in access control technology - be it for fraud detection or for something more mundane such as personalisation upon identity detection - competing methods were developed to address a need for robust, expressions-proof, and potentially uncertainty-aware (in the sense that degree of reliability can be reported) method of pairing a given, unseen 3-D scan with an entry in a database of faces (with unique preassigned IDs or equivalents). This document only deals with one family of approaches, namely those that non-rigidly transform 3-D data so as to score dissimilarity and therefore provide figures of merit to a given match. By trying to match many pairs or assessing their appropriateness for comparison based on a searching index5, one can determine a best match.
The following strands of work are most suitable for the proposed new framework and their appropriation will be described herein, where fusion of disparate ideas from each will be viewed as desirable for novelty.