Um estudo sobre reconhecimento facial 3D tolerante a oclusão e baseado em histogramas de atributos

Abstract

This study revolves around the context of individual recognition based on their 3D facial images. In order to identify individuals, local attribute description methods and relevant region location techniques are explored to allow classification even if immersed in a scenario where there are different types of occlusion. This is a challenging environment, since the occluded regions do not present a consistent match with the individual’s neutral face sample. The strategies applied to this study present the state-of-the-art results and have been tested on a commonly used database for facial recognition studies with partial occlusions. The experiments have achieved a recognition rate of 100% for neutral-neutral tests and 98.1%, 95.2%, 88.6% and 86.6% for tests utilizing samples with eye occlusion by hand, eye occlusion by glasses, mouth occlusion by hand and face occlusion by hair, respectively. All results correspond to rank-1 experiments using only a single training sample from the dataset (gallery) for each individual.