LEITE, B. B.; http://lattes.cnpq.br/2897008856605907; LEITE, Bruno de Brito.
Resumen:
There is a number of application scenarios in which eye detection plays a major role, such
as: human-computer interaction, facial feature tracking, driver vigilance monitoring, face
and iris recognition, facial expression analysis, and video conferencing systems. This dissertation presents a bibliographic review of recent eye detection related works. For each
reviewed work, a description is presented, containing the general approach adopted by the
authors, the achieved performance, and the image face database considered in the experiments (when information on this data is available). The main contribution of this dissertation is the proposition of a new approach for eye detection in face images. The pre-processing step includes illumination compensation using homomorphic filtering and histogram operations such as stretching and equalization for brightness and contrast improvement. The detection step uses: passive approaches based on (i) appearance analysis, regarding skin tone features; (ii) learning, which uses a neural network trained using features extracted from examples and counter examples of eyes; and (iii) template matching, which produces response from evaluation of template matching function. These approaches are integrated by three classifier combination rules (product, mean, and ranking). Experimental results are the proof of concept for the proposed approach. In these experiments, the performance of the proposed approach overcome the performance of existing approaches discussed in the literature review.