Consistent hierarchical labeling of image and image regions

Patent metadata

Field	Value
Publication number	US-9355337-B2
Application number	US-54694809-A
Country	US
Kind code	B2
Filing date	Aug 25, 2009
Priority date	Aug 25, 2009
Publication date	May 31, 2016
Grant date	May 31, 2016

Abstract

Classification of image regions comprises: recursively partitioning an image into a tree of image regions having the image as a tree root and at least one image patch in each leaf image region of the tree, the tree having nodes defined by the image regions and edges defined by pairs of nodes connected by edges of the tree; assigning unary classification potentials to nodes of the tree; assigning pairwise classification potentials to edges of the tree; and labeling the image regions of the tree of image regions based on optimizing an objective function comprising an aggregation of the unary classification potentials and the pairwise classification potentials.

First claim

The invention claimed is:

1. An image classifier comprising:
A digital processor configured to perform operations comprising:
recursively partitioning an image into a tree of image regions having the image as a tree root and at least one image patch in each leaf image region of the tree;
classifying the image regions using at least one classifier; and
outputting classification values for the image regions based on the classifying and on weights assigned to the nodes and edges of the tree.

2. The image classifier as set forth in claim 1 , wherein the output classification values are constrained by a hierarchy of classes.

3. The image classifier as set forth in claim 2 , wherein the constraint by a hierarchy of classes is applied by the weights assigned to edges of the tree.

4. The image classifier as set forth in claim 1 , wherein the digital processor is configured to further perform an operation comprising:
assigning labels to one or more image regions based on the output classification values.

5. The image classifier as set forth in claim 1 , wherein the digital processor is configured to further perform operations comprising:
repeating the recursive partitioning, classifying, and outputting for a plurality of images; and
retrieving one or more images of the plurality of images based on the output classification values for the image regions of the plurality of images.

6. The image classifier as set forth in claim 5 , wherein the digital processor is configured to further perform in cooperation with a display an operation comprising:
displaying the retrieved one or more images of the plurality of images.

7. The image classifier as set forth in claim 1 , wherein the at least one classifier includes a plurality of classifiers trained at different size scales, and the outputting comprises:
generating the output classification values by optimizing an objective function comprising an aggregation of:
(i) unary potentials defined by the classifying of image regions using classifiers trained at different size scales weighted by the weights assigned to the nodes of the tree and
(ii) pairwise potentials defined by pairs of the classifying of image regions using classifiers trained at different size scales connected by edges of the tree weighted by the weights assigned to the edges of the tree.

8. The image classifier as set forth in claim 1 , wherein the at least one classifier includes a plurality of classifiers, and the digital processor is configured to further perform operations comprising:
training the classifiers trained at different size scales using a training set of pre-classified image regions of different sizes; and
determining values of the weights assigned to the nodes and edges of the tree based on the classifiers trained at different size scales and the training set of pre-classified image regions of different sizes.

9. An image classification method comprising:
recursively partitioning an image into a tree of image regions having the image as a tree root and at least one image patch in each leaf image region of the tree, the tree of image regions having nodes defined by the image regions and edges defined by pairs of nodes connected by edges of the tree of image regions;
assigning unary classification potentials to nodes of the tree image regions;
assigning pairwise classification potentials to edges of the tree of image regions; and
labeling the image regions of the tree of image regions based on optimizing an objective function comprising an aggregation of the unary classification potentials and the pairwise classification potentials;
wherein the recursive partitioning, the assigning of unitary classification potentials, the assigning of pairwise classification potentials, and the labeling are performed by a digital processing device.

10. The image classification method as set forth in claim 9 , wherein the unary potentials are parameterized by a set of first-order classifiers.

11. The image classification method as set forth in claim 10 , further comprising:
training the set of first-order classifiers using a training set of pre-classified image regions; and
training parameters of the objective function using the same training set of pre-classified image regions;
wherein the recursive the training of the set of first-order classifiers and the training of parameters of the objective function are performed by the digital processing device.

12. The image classification method as set forth in claim 9 , wherein the unary potentials are parameterized by a set of classifiers trained at different size scales.

13. The image classification method as set forth in claim 9 , wherein the pairwise potentials encode inheritance constraints defined by a hierarchy of classes.

14. The image classification method as set forth in claim 9 , wherein the unary potentials are parameterized by a set of classifiers and the pairwise potentials encode inheritance constraints defined by a hierarchy of classes.

15. The image classification method as set forth in claim 14 , wherein the unary potentials are parameterized by a set of first-order classifiers trained at different size scales.

16. The image classification method as set forth in claim 9 , wherein the optimizing employs a Junction Tree Algorithm.

17. The image classification method as set forth in claim 9 , wherein the objective function can be decomposed into a sum of the unary classification potentials and the pairwise classification potentials.

18. A non-transitory storage medium storing instructions executable by a digital processor to perform an image classification method comprising:
recursively partitioning an image into a tree of image regions having the image as a tree root and at least one image patch in each leaf image region of the tree, the tree having nodes defined by the image regions and edges defined by pairs of nodes connected by edges of the tree;
assigning classification values to each node using a plurality of classifiers trained at different image size scales;
assigning a unary classification potential to each node based on the classification values assigned using the plurality of classifiers;
assigning a pairwise classification potential to each edge of the tree based on the classification values assigned using the plurality of classifiers trained at different image size scales and on inheritance constraints defined by a hierarchy of classes; and
labeling the image regions of the tree of image regions based on optimizing an objective function comprising an aggregation of the unary classification potentials and the pairwise classification potentials.

19. The non-transitory storage medium as set forth in claim 18 , wherein the assigning classification values comprises:
assigning classification values to each node using a plurality of classifiers trained at different image size scales.

20. The non-transitory storage medium as set forth in claim 18 , wherein the recursive partitioning comprises:
recursively partitioning the image into a quadtree of image regions.

Assignees

• Mcauley Julian
• De Campos Teofilo E
• Csurka Gabriela
• Perronnin Florent
• Xerox Corp

Inventors

• Mcauley Julian
• De Campos Teofilo E
• Csurka Gabriela
• Perronnin Florent

Classifications

• G06V10/7625

  Hierarchical techniques, i.e. dividing or merging patterns to obtain a tree-like representation; Dendograms · CPC title

• G06V10/809

  of classification results, e.g. where the classifiers operate on the same input data · CPC title

• G06F18/231

  Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram · CPC title

• G06V20/35 Primary

  Categorising the entire scene, e.g. birthday party or wedding scene · CPC title

• G06F18/254

  of classification results, e.g. of results related to same input data · CPC title