Camera Independent Face Recognition Algorithm In Visual Surveillance

Yew, Chuu Tian (2015) Camera Independent Face Recognition Algorithm In Visual Surveillance. Masters thesis, Universiti Sains Malaysia.

Preview

PDF - Submitted Version Download (2MB) | Preview

Abstract

Face recognition in visual surveillance has the ability to reduce crime rates in public area due to the suspect’s identity can be automatically identified in real-time using the face images captured by the surveillance camera as circumstantial evidence. Several available image preprocessing techniques, classifiers, and approaches had been proposed and tested to mitigate the effect of illumination variation, pose variations, and intensity quality differences due to hardware differences in such system. The face recognition system should be able to integrate seamlessly into the existing system. From the experiments, Histogram Equalization (HE) preprocessed face images scaled to 30�30 had proven to be well suited for pre-processing of surveillance images. The combination of Linear Discriminant Analysis (LDA) and HE preprocessed images managed to achieve an average recognition rate of 81.48% for the single camera training set. The flandmark facial landmark detector is implemented to determine the location of the eyes and new face images are obtained by cropping the HE pre-processed images. The combination of flandmark images at 20�30 with multi-class Support Vector Machine (SVM) is used to form a multimodal classification system with LDA and HE combination. Score level fusion is done to the normalized output scores of both the classifiers with proper weight, w assigned to each score. Finally, the watch list principle will list out several possible subjects according to their respective score ranking rather than deciding on a particular subject based on the maximum score, thus increasing the performance of the proposed system. The experimental results demonstrate the performance of the proposed algorithm on Surveillance Camera Face Database (SCface) database with 97.45% average recognition rate.

Item Type:	Thesis (Masters)
Additional Information:	Accession No: 875005805
Subjects:	T Technology > TK Electrical Engineering. Electronics. Nuclear Engineering > TK7870 Electronic packaging
Divisions:	Kampus Kejuruteraan (Engineering Campus) > Pusat Pengajian Kejuruteraaan Elektrik & Elektronik (School of Electrical & Electronic Engineering) > Thesis
Depositing User:	Mr Mohd Fadli Abd Rahman
Date Deposited:	10 Jul 2018 09:02
Last Modified:	10 Jul 2018 09:02
URI:	http://eprints.usm.my/id/eprint/40970

Actions (login required)

View Item