Construcción de un conjunto de imágenes faciales con metadatos biométricos y rasgos étnicos de ecuatorianos

Abstract

Currently, Machine Learning techniques are increasingly used in various areas of science and society, its operation is based on the systematic collection and analysis of data that allow create prediction models. Today, there are several sources to acquire data / images, such as public repositories of governments, academic institutions, research centers and data generated by companies, among others. Despite all these options, the adequate data for the construction of the models are not always available, especially when problems that require the use of images are addressed. This article proposes the construction of a set of facial images of Ecuadorians belonging to the most representative ethnic groups of Ecuador: afro-ecuadorians, mestizos, indigenous and european-descendants. In addition, metadata referring to the facial features of each ethnic group obtained through the application of a survey and the respective biometrics, calculated through the execution of ssd inception v2 coco and faster rcnn inception v2 coco, models of the TensorFlow library are included. The resulting image dataset has 430 instances, each with 24 attributes. The results in the detection and acquisition of the face biometrics were contrasted with physical measurements, achieving errors less than 10%.