Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ Recolector de Cienci...arrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
Biblos-e Archivo
Master thesis . 2019
Data sources: Biblos-e Archivo
versions View all 2 versions
addClaim

Incorporación de atributos faciales a sistemas de reconocimiento facial

Authors: Blázquez Cortés, Marta;

Incorporación de atributos faciales a sistemas de reconocimiento facial

Abstract

La aparición de redes neuronales profundas ha provocado un gran progreso en el ámbito de la biometría. Los sistemas de reconocimiento facial son cada vez más utilizados y cada vez requieren una mayor precisión. Un modo habitual de mejorar estos sistemas es el refuerzo mediante atributos característicos de cada persona, los llamados soft biometrics. El género, la edad o la raza son algunos de los atributos más habituales. Al analizar el rendimiento de los sistemas de reconocimiento facial se observan diferencias dentro de cada grupo demográ co. Atendiendo al género, las mujeres son las que peores resultados obtienen. Para el caso de la raza, son las personas de raza negra o asiática las que normalmente presentan más di cultades en el reconocimiento facial. Este problema radica principalmente en los conjuntos de entrenamiento con los que los modelos han aprendido. Estos no suelen estar balanceados y se re eja en los resultados cuando analizamos cada clase. Normalmente las bases de datos incluyen más hombres y más identidades de raza blanca. En este trabajo se desarrollan sistemas especí cos para los grupos demográ cos de género y raza. Los resultados experimentales demuestran que utilizando modelos entrenados con imágenes pertenecientes a una única clase se mejora el rendimiento de un sistema de reconocimiento facial genérico que ha sido entrenado con imágenes de todas las clases. Se proponen también dos estimadores para los atributos de género y raza. Se compara el rendimiento del sistema cuando la información de dichos atributos es obtenida de manera manual, es decir mediante etiquetas y cuando se extrae de manera automática. Además se propone un sistema más completo que fusiona la información de género y raza. Y se analizan las alternativas de fusión a nivel de features y a nivel de scores.

The research in deep neural networks has produced a great improvement in the world of biometrics. Facial recognition systems are used more often and require a higher accuracy. A common way of improving these systems is the reinforcement through characteristic attributes from each person which are known as soft biometrics. The gender, age or ethnic group are the most common attributes. Analyzing the performance of facial recognition systems, di erences are observed within each demographic group. Considering the gender, women obtain the worst results. Regarding the ethnicity group, dark skin persons or asian have more di culties in the facial recognition. This problem is mainly due to the training sets used for the learning process of the models. These are not usually balanced and that is re ected in the results obtained for each class. Usually datasets include more men and more white race identities. In this project, speci c models are developed for the demographic groups of gender and ethnicity. The experimental results show that using trained models with images from a single class, it is possible to improve the performance of a generic facial recognition system trained with images from all classes. Two estimators for the gender and ethnic group attributes are also proposed. System performance is compared when race and gender information is obtained automatically or manually, through label. Moreover, a more complete system is proposed combining gender and ethnic group information. Proposing a fusion of this information at the scores or the features level.

Máster Universitario en Investigación e Innovación en Tecnologías de la Información y las Comunicaciones

Country
Spain
Related Organizations
Keywords

Telecomunicaciones, Reconocimiento Facial, Reconocimiento Biométrico, Redes Neuronales Convolucionales

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green