Bags of visual context-dependent words for generic visual categorization

• US 8,165,410 B2
• Filed: 12/17/2010
• Issued: 04/24/2012
• Est. Priority Date: 09/19/2006
• Status: Active Grant

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Perronnin, Florent
• Primary Examiner(s): Liew, Alex
• Application Number:
  US12/971,087
• Publication Number:
  US 20110091105A1
• Time in Patent Office: 494 Days
• Field of Search: 382181-231
• US Class Current: 382/224
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]
• CPC Class Codes: G06V 10/464 using a plurality of salient features, e.g. bag-of-words [BoW] representations
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects

Abstract

Category context models (64) and a universal context model (62) are generated including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in training images (50) assigned to each category and assigned to all categories, respectively. Context information (76) about an image to be classified (70) are generated including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in the image to be classified. For each category (82), a comparison is made of (i) closeness of the context information about the image to be classified with the corresponding category context model and (ii) closeness of the context information about the image to be classified with the universal context model. An image category (92) is assigned to the image to be classified being based on the comparisons.

Claims

1. An image classification method comprising:

generating a category context model for each of a plurality of image categories, the category context model for an image category including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in training images assigned to the category;

generating context information about an image to be classified, the context information including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in the image to be classified; and

assigning an image category to the image to be classified based on comparison of the context information about the image with the category context models for the image categories;

wherein at least the generating of the category context model and the generating of context information are performed by a computing apparatus.

2. The image classification method as set forth in claim 1, further comprising:

generating a universal context model including sums of soft co-occurrences of pairs of visual words in geometric proximity to each other in training images assigned to any of the plurality of image categories; and

for each category, generating a comparison of (i) closeness of the context information about the image to be classified with the corresponding category context model and (ii) closeness of the context information about the image to be classified with the universal context model, the assigning of the image category to the image to be classified being based on the comparisons.

3. The image classification method as set forth in claim 1, wherein the soft co-occurrences are computed as products of at least occupation probability values for the visual words of the pair of visual words in neighboring patches of the training image or the image to be classified.

4. An image classifier comprising:

a vocabulary of visual words wherein a visual word is a defined grouping of low-level image features;

a patch context analyzer configured to generate a context representation for each of a plurality of patches of an image, wherein the context representation for a patch is based on occurrence probabilities of context visual words in a plurality of neighboring patches;

an image labeler configured to assign an image category to an image based at least on the context representations of a plurality of patches of the image, wherein the image labeler applies a category context model for each image category that indicates probabilities of context words being in a neighborhood of an occurrence of a vocabulary word for images of that image category; and

a category context model generator configured to generate each category context model as sums of soft co-occurrences of pairs of words in geometric proximity to each other in training images assigned to the category.

5. The image classifier as set forth in claim 4, wherein the context words consist of a sub-set of visual words of the vocabulary that is substantially smaller than the total number of words in the vocabulary.

6. The image classifier as set forth in claim 4, wherein the context words include all visual words of the vocabulary.

7. The image classifier as set forth in claim 4, wherein the context words include at least some derived words that are not included in the vocabulary of visual words but that are derived from visual words of the vocabulary.

8. The image classifier as set forth in claim 4, wherein the patch context analyzer generates the context representation for each patch as a plurality of values, each value being indicative of a statistical occupancy probability of a corresponding context word in neighboring patches weighted by closeness of the neighboring patches to the patch for which the context histogram is generated.

10. The image classifier as set forth in claim 4, wherein the image labeler includes a plurality of comparators each comparing (i) closeness of the context representations of the patches of the image with a category context model and (ii) closeness of the context representations of the patches of the image with a universal context model, the image category being assigned based on the outputs of the comparators.

9. An image classifier comprising:

a vocabulary of visual words wherein a visual word is a defined grouping of low-level image features;

a patch context analyzer configured to generate a context representation for each of a plurality of patches of an image, wherein the context representation for a patch is based on occurrence probabilities of context visual words in a plurality of neighboring patches;

an image labeler configured to assign an image category to an image based at least on the context representations of a plurality of patches of the image, wherein the image labeler applies (i) a category context model for each image category that indicates probabilities of context words being in a neighborhood of an occurrence of a vocabulary word for images of that image category and (ii) a universal context model that indicates probabilities of context words being in a neighborhood of an occurrence of a vocabulary word for images regardless of image category;

a category context model generator configured to generate each category context model as sums of soft co-occurrences of pairs of words in geometric proximity to each other in training images assigned to the category; and

a universal context model generator configured to generate the universal context model as sums of soft co-occurrences of pairs of words in geometric proximity to each other in training images assigned to all categories.

11. The image classifier as set forth in claim 9, wherein the context words consist of a sub-set of visual words of the vocabulary that is substantially smaller than the total number of words in the vocabulary.

12. The image classifier as set forth in claim 9, wherein the context words include all visual words of the vocabulary.

13. The image classifier as set forth in claim 9, wherein the context words include at least some derived words that are not included in the vocabulary of visual words but that are derived from visual words of the vocabulary.

14. The image classifier as set forth in claim 9, wherein the patch context analyzer generates the context representation for each patch as a plurality of values, each value being indicative of a statistical occupancy probability of a corresponding context word in neighboring patches weighted by closeness of the neighboring patches to the patch for which the context histogram is generated.

15. The image classifier as set forth in claim 9, wherein the image labeler includes a plurality of comparators each comparing (i) closeness of the context representations of the patches of the image with a category context model and (ii) closeness of the context representations of the patches of the image with a universal context model, the image category being assigned based on the outputs of the comparators.