Image based object classification

Patent metadata

Field	Value
Publication number	US-2016110630-A1
Application number	US-201314897916-A
Country	US
Kind code	A1
Filing date	Jun 13, 2013
Priority date	Jun 13, 2013
Publication date	Apr 21, 2016
Grant date	—

Abstract

A method for classifying an object in image data to one out of a set of classes using a classifier, said image data comprising an image of the object, each class indicating a property common to a group of objects, the method comprising the steps of obtaining said classifier used to estimate for an input feature vector a probability for each of the set of classes, one probability indicating whether the input feature vector belongs to one class; extracting a feature vector from said image data; using the obtained classifier to estimate the probabilities for the extracted feature vector; and evaluating the estimated probabilities for determining whether the object does not belong to any one of the set of classes based using a quality indicator.

First claim

1 . A method for classifying an object in image data to one out of a set of classes using a classifier, said image data comprising an image of the object, each class indicating a property common to a group of objects, the method comprising the steps of:
obtaining said classifier used to estimate for an input feature vector a probability for each of the set of classes, one probability indicating whether the input feature vector belongs to one class; extracting a feature vector from said image data; using the obtained classifier to estimate the probabilities for the extracted feature vector; and evaluating the estimated probabilities for determining whether the object does not belong to any one of the set of classes based using a quality indicator.

2 . The method of claim 1 , further comprising a step of generating said classifier, including extracting feature vectors from learning image data.

3 . The method of claim 2 , wherein the classifier is generated using a support vector machine.

4 . The method of claim 2 or 3 , further comprising:
separating template image data into said learning image data and into test image data;
testing the classifier including extracting feature vectors from said test image data; and
defining as said quality indicator a threshold probability based on the test result.

5 . The method of claim 4 , wherein the testing includes calculating probabilities that feature vectors extracted from the test image data are mapped to the correct classes, and defining the threshold probability as the smallest of the calculated probabilities.

6 . The method of claim 5 , wherein the testing is repeated n times using different test image data, and the threshold probability is defined as the smallest of the averaged calculated probabilities.

7 . The method of any one of claims 1 to 3 , wherein the quality indicator indicates entropy of the estimated probabilities.

8 . The method of any one of claims 1 to 7 , further comprising a step of adding a new class to the set of classes if it is determined that the object does not belong to any one of the set of classes.

9 . The method of any one of claims 1 to 8 , wherein the quality indicator is defined based on a reliability of said classifier.

10 . The method of any one of claims 1 to 9 , further comprising a step of counting a classified object.

11 . The method of any one of claims 1 to 10 , further comprising a step of authenticating a classified object.

12 . A device for classifying an object in image data, to one of a set of classes using a classifier, said image data comprising an image of the object, each class indicating a property common to a group of objects, the device comprising processing resources being configured to:
obtain said image data; obtain said classifier used to estimate for an input feature vector a probability for each of the set of classes, one probability indicating whether the input feature vector belongs to one class; extract a feature vector from said image data; use the obtained classifier to estimate the probabilities for the extracted feature vector; and to evaluate the estimated probabilities for determining whether the object does not belong to any one of the set of classes based using a quality indicator.

13 . The device of claim, wherein said processing resources are configured to implement a method of any one of claims 2 to 11 .

14 . The device of claim 12 or 13 , wherein the device further comprises image acquisition means for acquiring said image data.

15 . A computer program comprising code, said code, when being executed on a processing resource, implementing a method of any one of claims 1 to 11 .

16 . A computer program product comprising a tangible data carrier storing in a non-volatile manner the computer program of claim 15 .

17 . A system comprising:
a conveyor line arranged for moving objects through a field of view; the device of claim 12 or 13 ; and image acquisition means arranged for acquiring said image data comprising an image of said field of view;

18 . The system of claim 17 , wherein the conveyor line is any one of a conveyor belt, conveyor chain, guiding rail, sliding track, and transport track.

Assignees

• Sicpa Holding Sa

Inventors

• Heusch Guillaume
• Pican Nicolas

Classifications

• G06T7/0004 Primary

  Industrial image inspection · CPC title

• G06V10/765

  using rules for classification or partitioning the feature space · CPC title

• G06F18/28

  Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries · CPC title

• G06F18/2431

  Multiple classes · CPC title

• G06F18/2193

  based on specific statistical tests · CPC title