Browse Prior Art Database

Method and System for Spotting Celebrities in Professional Photo Libraries

IP.com Disclosure Number: IPCOM000214451D
Publication Date: 2012-Jan-30
Document File: 4 page(s) / 237K

Publishing Venue

The IP.com Prior Art Database

Related People

Sina Jafarpour: INVENTOR [+3]

Abstract

A method and system for spotting celebrities in professional photo libraries is disclosed. The method and system solves the problem of face recognition when face images vary in pose and illumination.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 49% of the total text.

Method and System for Spotting Celebrities in Professional Photo Libraries

Abstract

A method and system for spotting celebrities in professional photo libraries is disclosed.  The method and system solves the problem of face recognition when face images vary in pose and illumination.

Description

Disclosed is a method and system for spotting celebrities in professional photo libraries.  The professional photo libraries may correspond to a celebrity face repository containing face images of celebrities.  The method and system proposes a face recognition algorithm that deals with face images dynamically.  The face images may have variations in one or more of a pose and an illumination.  The face recognition algorithm uses a distance learning approach on pixel domain representation and a sparse approximation classifier for recognizing the face images.

Each face image in the face repository may be associated with one or more metadata such as name of a celebrity, title of the image, date when the face image was taken and location.  Initially, face detection is performed on the face image by detecting a face boundary.  Thereafter, histogram equalization may be applied to the face image to magnify contrast between face and background of the face image.

Figure 1 illustrates face images that were detected for plurality of celebrities after applying the histogram equalization.  The face images were detected under variation in the pose, the illumination and an occlusion.

Figure 1

Further, a subspace pursuit algorithm may be used for identifying face images that are closest to the face image.  The subspace pursuit algorithm is a greedy algorithm that is used for reconstruction of sparse and compressible signals in digital communications and radar.  The subspace algorithm has two important characteristics, low computational complexity and reconstruction accuracy of the same order as that of an l­1 minimization method.  Moreover, the subspace pursuit algorithm recovers a sparse vector even when the sparsity level is sufficiently small.  The face image may be represented as a pixel domain representation and a distance learning transformation may be applied on the pixel domain representation.  The distance learning transformation may be applied on the pixel domain representation for transforming the face image to a new domain.  The face image in the new domain represents the face image in a better way.  The distance learning transformation can be formalized as provided in equation (1),

minimize ………….(1)

subject to

where,

 = regularization parameter; and

= slack variables.

The distance learning transformation transforms face vectors, associated with the face image, from the pixel domain representation to a new transform domain.  Consequently, a new face repository is created in accordance with the face images in the new transform domain.  The distance learning transformation minimizes distance between the new face image and the face...