Detecção e reconhecimento de faces distorcidas por artefatos de compressão

Face recognition systems allow several applications, such as access control, biometric identification and surveillance. A study is presented on the resilience of facial detectors and facial regognition methods to faces distorted by compression artifacts. The JPEG, JPEG2000, H.264 / AVC and HEVC comp...

Full description

Bibliographic Details
Authors: Ferreira, Fernando Rodrigues Trindade, Pagliari, Carla Liberal
Format: article
Status:Published version
Publication Date:2018
Country:Brasil
Institution:Instituto Militar de Engenharia (IME)
Repository:C&T (Rio de Janeiro. Online)
Language:Portuguese
OAI Identifier:oai:meiramattos.eceme.ensino.eb.br:article/1957
Online Access:https://ebrevistas.eb.mil.br/CT/article/view/1957
Access Level:Open access
Keyword:Codification
face recognition
face detection
coding artifacts
Codificação
reconhecimento de faces
detecção de faces
artefatos de codificação
Description
Summary:Face recognition systems allow several applications, such as access control, biometric identification and surveillance. A study is presented on the resilience of facial detectors and facial regognition methods to faces distorted by compression artifacts. The JPEG, JPEG2000, H.264 / AVC and HEVC compression standards are used. For face recognition, two algorithms are studied: LBP and PCA. This work proposes a combination of two versions of the Viola-Jones detector to mitigate the occurrence of false positives. Another contribution of this work is the use of original and coded sequences of images / videos, as well as the use of traditional face datasets for face detection and recognition. The results show that the LBP method outperforms the PCA method, as the LBP is more robust to the changes imposed by the coding artifacts since it analyzes the textures of the images