Open Access System for Information Sharing

Login Library

 

Article
Cited 1 time in webofscience Cited 1 time in scopus
Metadata Downloads

An accurate and real-time multi-view face detector using ORFs and doubly domain-partitioning classifier

Title
An accurate and real-time multi-view face detector using ORFs and doubly domain-partitioning classifier
Authors
Yoon, J.Kim, D.
Date Issued
Dec-2019
Publisher
SPRINGER HEIDELBERG
Abstract
We propose a novel multi-view face detector that operates accurately and fast in challenging environments. It consists of four consecutive functional components: background rejector, pose classifier, pose-specific face detectors, and face validator. The background rejector removes non-face patches quickly, the pose classifier estimates poses of the surviving patches, one or more selected pose-specific face detectors according to their estimated pose labels determine that a given patch is a face by using winner take all (WTA) strategy, and the face validator checks whether the face-like patch is really a face. For achieving strong discrimination power with low computing overhead, we devise several types of order relation features (ORF) that encode the order relation among feature elements as a unique code. The devised ORFs are placed in functional components appropriately to ensure fast operation of the multi-view face detector. For accurate classification, we propose a doubly domain-partitioning (DDP) classifier that consists of a coarse domain-partitioning weak classifier followed by a fine bin-partitioning weighted linear discriminant analysis (wLDA) classifier. For fast classification, we devise a feature sharing method that shares identical features between the background rejector and the pose classifier, and among all classes in the pose classifier. We evaluated the proposed multi-view face detector using the FDDB, AFW, and PASCAL face datasets. The experimental results show that the proposed multi-view face detector outperforms other state-of-the-art methods in terms of detection accuracy and execution time.
URI
http://oasis.postech.ac.kr/handle/2014.oak/101186
ISSN
1861-8200
Article Type
Article
Citation
JOURNAL OF REAL-TIME IMAGE PROCESSING, vol. 16, no. 6, page. 2425 - 2440, 2019-12
Files in This Item:
There are no files associated with this item.

qr_code

  • mendeley

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Related Researcher

Researcher

김대진KIM, DAI JIN
Dept of Computer Science & Enginrg
Read more

Views & Downloads

Browse