Method and system for learning object recognition in images

• US 20080118105A1
• Filed: 11/16/2006
• Published: 05/22/2008
• Est. Priority Date: 11/16/2006
• Status: Active Grant

• Current Assignee: Innovation Asset Collective
• Original Assignee: Tandent Vision Science, Inc.
• Inventors: Friedhoff, Richard Mark, Maxwell, Bruce Allen
• Granted Patent: US 7,853,071 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06T 7/90 Determination of colour characteristics
• CPC Class Codes: G06V 10/60 relating to illumination properties, e.g. using a reflectance or lighting model

Abstract

In a first exemplary embodiment of the present invention, an automated, computerized method for learning object recognition in an image is provided. According to a feature of the present invention, the method comprises the steps of providing a training set of standard images, calculating intrinsic images corresponding to the standard images and building a classifier as a function of the intrinsic images.

Claims

1. An automated, computerized method for learning object recognition in an image, comprising the steps of:

providing a training set of standard images;

calculating intrinsic images corresponding to the standard images; and

building a classifier as a function of the intrinsic images.

2. The method of claim 1 comprising the further step of utilizing the classifier to identify objects of interest in the image.

3. The method of claim 2 wherein the image comprises a standard image and a corresponding intrinsic image.

4. The method of claim 1 wherein the step of building a classifier as a function of the intrinsic images is carried out to build a classifier utilizing a computer learning technique selected from the group consisting of a feature-based learning framework, an appearance-based learning framework, shape-based analysis and classification, and model-based object recognition.

5. The method of claim 1 wherein the step of calculating intrinsic images corresponding to the standard images is carried out by providing a standard image comprising an array of pixels, each pixel having N band color values;

generating a histogram on a bi-illuminant chromaticity plane, the histogram being divided into bins;

calculating an envelope value for each bin of the histogram;

co-relating each of the pixels of the standard image to a corresponding bin of the histogram; and

mapping the N band color values of each pixel to the envelope value of the co-related bin to provide a material image.

6. The method of claim 5 comprising the further step of utilizing the material image to generate an illumination image.

7. The method of claim 1 wherein the classifier comprises an Adaboost classifier.

8. The method of claim 1 wherein the intrinsic images comprise material images and illumination images.

9. The method of claim 8 wherein the step of building a classifier as a function of the intrinsic images is carried out by selecting image features from each of the material images and illumination images.

10. The method of claim 9 comprising the further step of selecting image features as a function of relationships between features from each of the material images and illumination images.

11. The method of claim 10 wherein the relationships between features from each of the material images and illumination images comprise features selected from the group consisting of differences and sums of features, products and ratios of features, sums of products, and products of sums.

12. The method of claim 1 wherein the step of calculating intrinsic images corresponding to the standard images is carried out by providing a BIDR model representation of the standard images, and generating the corresponding intrinsic images as a function of the BIDR model representation.

13. An automated, computerized method for recognizing an object in an image, comprising the steps of:

generating a classifier; and

utilizing the classifier to identify objects of interest in an image comprising a standard image and a corresponding intrinsic image.

14. An automated, computerized method for recognizing an object in an image, comprising the steps of:

providing a classifier;

providing an image;

calculating an intrinsic image corresponding to the image; and

utilizing the classifier to identify objects of interest in the intrinsic image.

15. The method of claim 14 wherein the step of calculating an intrinsic image corresponding to the image is carried out by providing a BIDR model representation of the standard images, and generating the corresponding intrinsic images as a function of the BIDR model representation.

16. A computer system which comprises:

a CPU; and

a memory storing image files, each defining an image comprising an array of pixels;

the CPU being arranged and configured to execute a routine;

to provide a training set of standard images;

to calculate intrinsic images corresponding to the standard images; and

to build a classifier as a function of the intrinsic images.

17. A computer system which comprises:

a CPU; and

a memory storing image files, each defining an image comprising an array of pixels;

the CPU being arranged and configured to execute a routine;

to generate a classifier; and

to utilize the classifier to identify objects of interest in an image comprising a standard image and a corresponding intrinsic image.

18. A computer system which comprises:

a CPU; and

a memory storing image files, each defining an image comprising an array of pixels;

the CPU being arranged and configured to execute a routine;

to provide a classifier;

to provide an image;

to calculate an intrinsic image corresponding to the image; and

to utilize the classifier to identify objects of interest in the intrinsic image.

19. An automated, computerized method for recognizing an object in an image, comprising the steps of:

providing a classifier generated as a function of intrinsic images derived from a standard image training set and selected from the group consisting of a material image and an illumination image; and

utilizing the classifier to identify objects of interest in an image.

20. The method of claim 19 wherein the step of providing a classifier generated as a function of intrinsic images derived from a standard image training set and selected from the group consisting of a material image and an illumination image is carried out by generating a classifier as a function of features from the intrinsic images selected from the group consisting of features calculated from the material image, features calculated from the illumination image, and features calculated from relationships between both the material image and the illumination image.