Automatic ear detection and feature extraction using Geometric Morphometrics and Convolutional Neural Networks

CINTAS, CELIA; QUINTO-SÁNCHEZ MIRSHA; ACUÑA, VICTOR; PASCHETTA, CAROLINA ANDREA; DE AZEVEDO, SOLEDAD; SILVA DE CERQUEIRA, CAIO; RAMALLO, VIRGINIA; GALLO, CARLA; POLETTI, GIONANNI; BORTOLINI, MARIA CÁTIRA; CANIZALES-QUINTEROS, SAMUEL; ROTHHAMMER, FRANCISCO; BEDOYA, GABRIEL; RUIZ-LINARES, ANDRES; GONZÁLEZ-JOSÉ, ROLANDO; DELRIEUX, CLAUDIO

IET Biometrics

IET Digital Library

Año: 2016

2047-4938

Accurate gathering of phenotypic information is a key aspect in several subject matters, includingbiometric identification, biomedical analysis, forensics, and many other. Automatic identificationof anatomical structures of biometric interest, such as fingerprints, iris patterns, or facial traits,are extensively used in applications like access control, anthropological research, and surveillance,all having in common the drawback of requiring intrusive means for acquiring the required information.In this regard, the ear structure has multiple advantages. Not only the ear?s biometricmarkers can be easily captured from the distance with non intrusive methods, but also they experimentalmost no changes over time, and are not influenced by facial expressions. However, the mostrelevant proposals in the literature for automated ear detection do not take advantage of its phenotypicattributes that biologically determine the ear?s shape and form. Therefore, these methodsrequire acquisition in controlled conditions to perform adequately.In this paper we present a new method, based on two well established methodologies, GeometricMorphometrics and Deep Learning algorithms, for automatic ear detection and feature extractionin the form of 2D landmarks. A convolutional neural network was trained with a set of manuallylandmarked examples. The trained network is able to provide morphometric landmarks on ears?images automatically, with a performance that matches human assisted landmarking. The abilityto perform in the open (i.e., in images or video taken with no specific acquisition preparation),together with the feasibility of using ear landmarks as feature vectors for people identification,opens a novel spectrum of biometrics applications.