Abstract
Face recognition is a biometric measure that identifies a person through images of their face. Generally, a complete face recognition system must first detect the faces in a given image. The detected faces must then be represented in a format that allows a similarity metric to be calculated that can be compared to known faces stored in a database. Eigenfaces is a common face recognition method that represents faces through a weighted set of eigenvectors while measuring the similarity between two faces as the Euclidean distance between two weight vectors. Immune inspired anomaly detection identifies samples that do not conform to normal behaviour using methods inspired by the immune system’s ability to detect and react to antigens. The objective of this dissertation is to identify the potential advantages of applying immune inspired anomaly detection to the problem of face recognition. To this end, an environment simulator component provides input to the proposed model that mimics common face recognition scenarios. The proposed model is called the Immune Inspired Symbiotic Face Recognition Agents Model (IISFRAM). A proof of concept prototype called the Immune Inspired Symbiotic Face Recognition Agents Prototype (IISFRAP), will be implemented. The conceptual model and the proof of concept prototype consist of three intelligent agents that work together to provide a complete face recognition solution. The relationships between the agents are inspired by symbiotic relationships in nature. Each agent’s role in the IISFRAM is inspired by a component of the humoral immune system. Two scenarios that showcase the advantages of the proposed model are social media applications and physical access control. In both cases, the proposed model maintains an updated detector set, provides distributed processing, and modular components that allow any specific algorithm to be easily replaced. Testing the IISFRAP showed that a small detector set (26 detector agents) can completely cover a larger training set (90 samples) while successfully differentiating between new self-samples and non-self-samples. The IISFRAP had improved accuracy compared to similar systems with the immune inspired detector agents allowing for precision and recall metrics to be increased...
M.Sc. (Information Technology)