Clustering appearances of objects under varying illumination conditions

Claim 2

In a set of digital images, each representing a subject having an identity and each digital image having a corresponding illumination condition, wherein a subset of the set of digital images representing a same corresponding subject are related by a mathematical representation of a structure in a multidimensional space, a computer based method for identity based clustering the digital images having various illumination conditions and representing a first subject having a first identity, the method comprising: composing a set of affinity measures for every pair of digital images in the set of digital images, each affinity measure representative of a relationship between the pair of digital images of the set of images with respect to the structure in the multidimensional space; and applying an identity-based clustering algorithm to the set of affinity measures to determine a first subset of the set of images representing the first subject, wherein the images of said first subset are related by the mathematical representation of the structure in the multidimensional space, and wherein the first identity is undetermined prior to said step of applying the identity-based clustering algorithm.

Claim 3

The computer based method of claim 1 , further comprising the step of applying a K-subspace clustering to a result of the clustering algorithm to increase accuracy of the identification of the first subset of images.

Claim 4

The computer based method of claim 1 , wherein the clustering algorithm is a spectral clustering method.

Claim 5

The computer based method of claim 1 , wherein the structure in the multidimensional space is a convex polyhedral cone.

Claim 6

The computer based method of claim 4 , wherein the affinity measures are a set of non-negative coefficients of a linear least square approximation of each digital picture, each digital picture approximated as a summation of a linear combination of all other digital pictures in the set of digital pictures.

Claim 7

The computer based method of claim 5 , wherein the affinity measure is substantially zero for any two digital images representing different subjects.

Claim 8

The computer based method of claim 5 , wherein the set of affinity measures is represented by an affinity matrix.

Claim 9

The computer based method of claim 4 , wherein the affinity measures are a set of non-negative coefficients of a linear least square approximation of each digital picture, each digital picture approximated as a summation of a linear combination of a second subset of closest neighboring digital pictures in the set of digital pictures.

Claim 10

The computer based method of claim 1 , wherein the subjects are human faces.

Claim 11

The method of claim 1 , wherein each affinity measure is non-negative.

Claim 12

The method of claim 1 , wherein the identity-based clustering algorithm is based on subject identity.

Claim 13

In a set of digital images, each representing a subject having an identity and each digital image having a corresponding illumination condition, wherein a subset of the set of digital images representing a same corresponding subject are related by a mathematical representation of a structure in a multidimensional space, a computer system for identity based clustering the digital images having various illumination conditions and representing a first subject having a first identity, the system comprising: means for composing a set of affinity measures for every pair of digital images in the set of digital images, each affinity measure representative of a relationship between the pair of digital images of the set of images with respect to the structure in the multidimensional space; and means for applying an identity-based clustering algorithm to the set of affinity measures to determine a first subset of the set of images representing the first subject, wherein the images of said first subset are related by the mathematical representation of the structure in the multidimensional space, and wherein the first identity is undetermined prior to said applying the identity-based clustering algorithm.

Claim 14

In a set of digital images, each representing a subject having an identity and each digital image having a corresponding illumination condition, wherein a subset of the set of digital images representing a same corresponding subject are related by a mathematical representation of a structure in a multidimensional space, a computer system for identity based clustering the digital images having various illumination conditions and representing a first subject having a first identity, the system comprising: means for composing a set of affinity measures for every pair of digital images in the set of digital images, each affinity measure representative of a relationship between the pair of digital images of the set of images with respect to the structure in the multidimensional space; and means for applying an identity-based clustering algorithm to the set of affinity measures to determine a first subset of the set of images representing the first subject, wherein the images of said first subset are related by the mathematical representation of the structure in the multidimensional space, and wherein the first identity is undetermined prior to said applying the identity-based clustering algorithm.

Claim 15

The computer system of claim 12 , wherein each affinity measure is non-negative.

Claim 16

An image processing computer system for identifying images representing subjects comprising: an input module for receiving data representative of a set of digital images, each digital image representative of one subject having an identity; a memory device coupled to the input module for storing the data representative of the set of digital images; a processor coupled to the memory device for iteratively retrieving data representative of two images of the set of images, the processor configured to calculate a conic affinity measure between the two digital images, configured to compose an affinity matrix with the conic affinity measures of substantially all the digital images of the set of digital images, and configured to apply an identity-based clustering algorithm to the affinity matrix to determine a subset of digital images corresponding to a same subject having a first identity, wherein the images of said subset are related by a mathematical representation of a structure in a multidimensional space, and wherein the first identity is undetermined prior to said applying the identity-based clustering algorithm.

Claim 17

An image processing computer system for identifying images representing subjects comprising: an input module for receiving data representative of a set of digital images, each digital image representative of one subject having an identity; a memory device coupled to the input module for storing the data representative of the set of digital images; a processor coupled to the memory device for iteratively retrieving data representative of two images of the set of images, the processor configured to calculate a conic affinity measure between the two digital images, configured to compose an affinity matrix with the conic affinity measures of substantially all the digital images of the set of digital images, and configured to apply an identity-based clustering algorithm to the affinity matrix to determine a subset of digital images corresponding to a same subject having a first identity, wherein the images of said subset are related by a mathematical representation of a structure in a multidimensional space, and wherein the first identity is undetermined prior to said applying the identity-based clustering algorithm.

Claim 18

The computer system of claim 14 , wherein said conic affinity measure is non-negative.