Agents based Context-Aware Framework for Facial Identification System

Research in face recognition systems has been focused on improving algorithm performance under specific conditions or to increase the average performance under heterogeneous conditions. Typical systems do not adjust well to perform at optimal level for a given instant of identification because the algorithms, face feature representations are fixed to achieve the best average result. Therefore, there is a real need to design a context-aware adaptive face identification system that can select the best pre-processing, features, and classifier for any given instance of identification. This paper focuses on the practical implementation and evaluation of our proposed framework [1] [2] that is aware of its operational context and adapt itself to select a suitable approach to identify a given face image. This is by using agent technology to give the system an intelligent and adaptive mechanism to make decisions at the key stages of the facial identification process. The agents will use context information such as environment conditions and application requirements to select the most appropriate pre-processing, features and match scores to optimise the best identification accuracy for a given test image. Within our framework, we propose the use of agents in two strategies: 1) an agent-based adaptive score selection technique as an alternative to the traditional fusion approaches, and 2) an agent-based integrated technique as an improvement to the existing adaptive and non-adaptive techniques. The experimental results presented here demonstrate that our techniques of using agents outperform the traditional fusion strategy that is commonly used in face recognition systems as well as the performance of other existing techniques.