Efficient adaptive face recognition systems based on capture conditions

摘要: In many face recognition (FR) applications, changing capture conditions lead to divergence between facial models stored during enrollment and faces captured operations. Moreover, it is often costly or infeasible several high quality reference samples a priori design representative models. Although self-updating using high-confidence captures appear promising, they raise challenges when change. particular, of individuals may be corrupted by misclassified input captures, their growth require pruning bound system complexity over time. This paper presents for self-update that exploits changes in assure the relevance templates limit template galleries. The set (facial model) an individual only updated include new are under significantly different conditions. particular implementation this system, illumination detected order select from bio-login gallery. Face built-in still video camera taken at periodic intervals authenticate user having accessed secured computer network. Experimental results produced with DIEE dataset show proposed provides comparable level performance FR self-updates gallery on all but lower complexity, i.e., number per individual.