Face recognition has been an intensely active area of research over the past 25 years, yet methods based on ordinality are a fairly recent innovation. In this paper we demonstrate the advantages of these methods and then introduce innovations which greatly improve recognition accuracy, speed and stability. The innovations are automatic parameter selection based on training database characteristics, weighted sub-windows to increase the effect of more salient regions, component-based recognition for greater speed and accuracy, and a 'divide and conquer' technique applied at the final recognition stage to optimise separability of the database images and further reduce run times. Robustness to pose, illumination and expression variations are evaluated, and we expand the ordinal method to use colour and 3-dimensional images.
展开▼