E-Locus - Institutional Repository of the University of Crete

Home Collections School/Department School of Sciences and Engineering Department of Computer Science Post-graduate theses

Post-graduate theses

Current Record: 711 of 824

[Add to Basket]

Identifier

uch.csd.msc//1998georgezikos

Title

Ανίχνευση και Αναγνώριση Προσώπων με Χρήση της Ανάλυσης Wavelet

Alternative Title

Α Wavelet Based Framework For Human Face Detection and Recognition

Creator

Zikos, George Th

Abstract

Since a few years, content-based indexing methods are becoming of great interest for image and video retrieval in audio-visual archives. Detecting and recognizing human faces automatically in video data provide users with powerful tools for performing queries. In this work, a new scheme for face detection and recognition using a wavelet packet decomposition is presented. For face recognition, there is a preprocessing step where the face region is extracted from image background. After that every face is described by a subset of band filtered images containing wavelet coefficients. These coefficients characterize the face texture and a set of simple statistical measures permits to form compact and meaningful feature vectors. Then, an efficient and reliable probabilistic metric derived from the Bhattacharyya distance is used in order to classify the face feature vectors. Distances between the vector of the input face and the vectors of the faces in the face database are computed. The closest person to the input face is the one with the smallest associated distance. For detection, skin-tone color statistics are used in a preprocessing step for the extraction of candidate face regions. Each of these regions is described, as for recognition, by a subset of band filtered images containing the wavelet coefficients. The characteristics are extracted as in the face recognition case and the Bhattacharyya distance between the vectors of all the candidate face regions and the training vectors are computed. Based on this distance every region is classified as face or non face region. In both stages of color and vector classification a domain knowledge extracted from training data is used.

Issue date

1998-11-01

Date available

1998-11-24

Collection

School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses

Type of Work--Post-graduate theses

Permanent Link

https://elocus.lib.uoc.gr//dlib/a/a/6/metadata-dlib-1998georgezikos.tkl

Views

428

Digital Documents
	Download document View document Views : 7
	View document