Citas bibligráficas
Bautista, D., Miñano, P. (2019). Análisis de la influencia del ángulo lumínico sobre el rendimiento de las técnicas de reconocimiento facial basadas en la apariencia lineal [Tesis, Universidad Privada Antenor Orrego - UPAO]. https://hdl.handle.net/20.500.12759/5555
Bautista, D., Miñano, P. Análisis de la influencia del ángulo lumínico sobre el rendimiento de las técnicas de reconocimiento facial basadas en la apariencia lineal [Tesis]. PE: Universidad Privada Antenor Orrego - UPAO; 2019. https://hdl.handle.net/20.500.12759/5555
@misc{renati/381369,
title = "Análisis de la influencia del ángulo lumínico sobre el rendimiento de las técnicas de reconocimiento facial basadas en la apariencia lineal",
author = "Miñano Pereira, Paulo Sergio",
publisher = "Universidad Privada Antenor Orrego - UPAO",
year = "2019"
}
This research work proposes the analysis of the influence of the light angle on the performance of facial recognition techniques based on linear appearance in order to determine the degree of assertiveness of each technique at different angles of light. The study corresponds to an applied and quasi-experimental investigation; whose unit of analysis is the capture of faces with angular variations of illumination in the coronal plane. The methodology developed includes: documentary collection and research into recognition techniques based on linear appearance most commonly used in embedded systems; the development of each recognition system, the implementation of base algorithms for each technique, and the taking of photographs of the faces of collaborators with different lighting angles. For the development of the PCA technique, two algorithms were developed, the first: in charge of processing and reorganizing the data matrix of the photos and the second: in charge of the recognition task, whose task is to find the Euclidean distances and determine the face that had a higher percentage of matching facial features. For the LDA technique, the Fisher, FLD (Fisher Linear Discriminant) algorithm was used. For LPP, given that its treatment acquires greater complexity, it was decided to develop two complementary codes and thus develop its operation. Finally, a particular algorithm was used to obtain the Euclidean distances and, in turn, to be able to subsequently execute the recognition of said technique. MatLab was used for the implementation, taking advantage of its GUIDE tool for the development of a simple operation platform that can operate autonomously. The results shown in comparative graphs between the yields of the PCA, LDA and LPP techniques, on the photographs under the influence of the different light angles show that, for illumination angles superior to 20 ° of the horizontal axis of the face, the degradation of the efficiency it is at least 40% in the three techniques used.
Este ítem está sujeto a una licencia Creative Commons Licencia Creative Commons