Method of computing global-to-local metrics for recognition
A method of computing global-to-local metrics for recognition. Based on training examples with feature representations, the method automatically computes a local metric that varies over the space of feature representations to optimize discrimination and the performance of recognition systems. Given a set of points in an arbitrary features space, local metrics are learned in a hierarchical manner that give low distances between points of same class and high distances between points of different classes. Rather than considering a global metric, a class-based metric or a point-based metric, the proposed invention applies successive clustering to the data and associates a metric to each one of the clusters.
1. A method for global-to-local metric learning for classification and recognition, the method comprising:
in response to a set of hierarchically clustered points {x i , I=1 . . . N}, iteratively performing, by a processor, following operations:
performing a global metric learning operation on the set of points to estimate a global metric, wherein the set of hierarchically clustered points is represented—using a tree structure constructed with a clustering algorithm at each level,
for each of the points {x i , I=1 . . . N}, performing a transformation using a corresponding transformation matrix to generate transformed points {y i,j =π k=0 i−1 A i−1,j /Kx i,j }, wherein K represents a number of clusters, wherein A i,j a transformation matrix, and
clustering, using a clustering algorithm, the transformed points to generate a metric tree, wherein the global metric learning operation and transformation are performed until a termination criterion is satisfied, which is one of a maximum height in the metric tree, a minimal variance of data points in the metric tree, and a minimum number of data points the metric tree; and
using the metric tree to evaluate an image for pattern recognition.
2. The method according to claim 1 wherein said clustering algorithm is the K-means clustering.
3. The method according to claim 1 wherein said metric is a symmetric matrix obtained with the ITML algorithm.
4. The method according to claim 1 wherein said clustering algorithm uses the local metric at each node.
5. A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform a method for global-to-local metric learning, the method comprising:
and recognition, the method comprising:
in response to a set of hierarchically clustered points {x i , I=1 . . . N}, iteratively performing, by a processor, following operations:
performing a global metric learning operation on the set of points to estimate a global metric, wherein the set of hierarchically clustered points is represented using a tree structure constructed with a clustering algorithm at each level,
for each of the points {x i , I=1 . . . N}, performing a transformation using a corresponding transformation matrix to generate transformed points {y i,j =π k=0 i−1 A i−1,j /Kx i,j }, wherein K represents a number of clusters, wherein A i,j represents a transformation matrix, and
clustering, using a clustering algorithm, the transformed points to generate a metric tree, wherein the global metric learning operation and transformation are performed until a termination criterion is satisfied, which is one of a maximum height in the metric tree, a minimal variance of data points in the metric tree, and a minimum number of data points the metric tree; and
using the metric tree to evaluate an image for pattern recognition.
6. A medium according to claim 5 , wherein the clustering algorithm is the K-means clustering.
7. A system, comprising:
a processor; and
a memory coupled to the processor for storing instructions, which when executed from the memory, cause the processor to perform operations, the operations including
in response to a set of hierarchically clustered points {x i , I=1 . . . N}, iteratively performing, by a processor, following operations:
performing a global metric learning operation on the set of points to estimate a global metric, wherein the set of hierarchically clustered points is represented using a tree structure constructed with a clustering algorithm at each level,
for each of the points {x i , I=1 . . . N}, performing a transformation using a corresponding transformation matrix to generate transformed points {y i,j =π k=0 i−1 A i−1,j /Kx i,j }, wherein K represents a number of clusters, wherein A i,j represents a transformation matrix, and
clustering, using a clustering algorithm, the transformed points to generate a metric tree, wherein the global metric learning operation and transformation are performed until a termination criterion is satisfied, which is one of a maximum height in the metric tree, a minimal variance of data points in the metric tree, and a minimum number of data points the metric tree; and
using the metric tree to evaluate an image for pattern recognition.
8. A system according to claim 7 where the feature representations represent objects in images.
9. A system according to claim 8 where the objects are faces.
10. The medium according to claim 5 wherein said metric is a symmetric matrix obtained with the ITML algorithm.
11. The medium according to claim 5 wherein said clustering algorithm uses the local metric at each node.
12. The system according to claim 7 , wherein the clustering algorithm is the K-means clustering.
13. The system according to claim 7 wherein said metric is a symmetric matrix obtained with the ITML algorithm.
14. The system according to claim 7 wherein said clustering algorithm uses the local metric at each node.