CLASSIFICATION SYSTEM FOR SIMILAR OBJECTS FROM DIGITAL IMAGES

• US 20160300123A1
• Filed: 12/01/2014
• Published: 10/13/2016
• Est. Priority Date: 12/01/2013
• Status: Active Grant

• Current Assignee: Sky Alibhai, Zoe Jewell
• Original Assignee: Wildtrack
• Inventors: JEWELL, Zoe, ALIBHAI, Sky
• Granted Patent: US 9,911,066 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/23 Clustering techniques
• CPC Class Codes: G06F 18/285 Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
• CPC Class Codes: G06F 3/04842 Selection of displayed objects or displayed text elements G06F3/0482 takes precedence

Abstract

Aspects of the present disclosure may include a method of classifying an object that includes, receiving a set of images of an indication of an object, extracting data from the images, performing a comparison of the extracted image data to a set of comparison data associated with a set of comparative objects, determining a classification of the object based on the comparison and outputting data representative of the classification. The method may classify an object by an individual identity or by one or more group characteristics.

Claims

1. A method of classifying an object comprising:

receiving a set of images of an indication of the object;

extracting image data from each image comprising the set of images of the indication of the object;

comparing the image data to a set of comparison data, wherein the set of comparison data comprises a plurality of data sets, wherein each of the plurality of data sets is associated with a comparative object;

determining a classification of the object based on the comparison; and

outputting, for display, output data representative of the determined classification of the object.

2. The method of claim 1, wherein extracting image data comprises:

providing a plurality of landmark points on an image of the set of images of the indication of the object;

providing a plurality of derived points on the image;

generating object data, wherein object data comprises at least one of an area, an angle or a distance between landmark points and/or derived points; and

outputting the object data.

3. The method of claim 2, wherein extracting image data further comprises adjusting the resolution and orientation of the image to substantially match the resolution and orientation of a sample object image.

4. The method of claim 3, wherein the sample object image comprises a number of sample landmark points, and wherein each landmark point provided on the image corresponds to a corresponding sample landmark point on the sample object image.

5. The method of claim 2, wherein a position of each derived point is determined by the positions of a number of landmark points.

6. The method of claim 2, wherein extracting image data further comprises providing two scale points on the image, positioned with respect to a measuring reference depicted in the image such that the positions of the two scale points correspond to a reference distance of the measuring reference.

7. The method of claim 1, wherein comparing the image data set to a set of comparison data comprises:

generating, through discriminant analysis of the image data and the set of comparison data, an image set of canonical variates corresponding to the image data, a plurality of comparison sets of canonical variates, wherein each of the plurality of comparison sets of canonical variates corresponds to a comparative object, and an RCV set of canonical variates corresponding to a cumulative set of image data and comparison data;

generating a first centroid value corresponding to the image set of canonical variates, a plurality of comparison centroid values, wherein each of the plurality of comparison centroid values corresponds to a comparison set of canonical variates of the plurality of comparison sets of canonical variates, and an RCV centroid value corresponding to the RCV set of canonical variates;

plotting, in two-dimensional space, the first set of canonical variates as a first canonical centroid plot represented by a first ellipse having a center point at the first centroid value, one of the plurality of comparison sets of canonical variates as a second canonical centroid plot represented by a second ellipse having a center point at one of the plurality of comparison centroid values, and the RCV set of canonical variates as an RCV canonical centroid plot, represented by a third ellipse having a center point at the RCV centroid value; and

determining whether the first ellipse overlaps the second ellipse.

8. The method of claim 7, wherein determining a classification of the object comprises:

classifying the object as having the same individual identity as a comparative object in response to determining that the first ellipse overlaps the second ellipse.

9. The method of claim 8 wherein the output data representative of the determined classification of the object comprises an indication that the object has the same individual identity as the comparative object.

10. The method of claim 7, wherein comparing the image data set to a set of comparison data further comprises:

determining a plurality of distance values wherein a distance value comprises the distance between the first centroid value and a comparison centroid value of the plurality of comparison centroid values, the first centroid value and the RCV centroid value, a comparison centroid value of the plurality of comparison centroid values and the RCV value, or a first comparison centroid value of the plurality of comparison centroid values and a second comparison centroid value of the plurality of comparison centroid values;

applying a clustering technique to the plurality of distance values; and

generating, responsive to applying the clustering technique, a cluster dendrogram.

11. The method of claim 10, wherein the clustering technique comprises Ward'"'"'s clustering technique.

12. The method of claim 10, wherein the output data representative of the determined classification of the object comprises a cluster dendrogram.

13. The method of claim 1, wherein comparing the image data set to a set of comparison data comprises:

performing a validated discriminant analysis of the image data, comprising;

identifying a plurality of sets of known object data, wherein each of the plurality of sets of known object data comprises one of the plurality of sets of data which comprise the comparison data;

receiving a selection of variables; and

responsive to receiving the selection of variables, generating, through discriminant analysis, a canonical plot of the image data and the plurality of sets of known object data, wherein the image data is located at a location in the canonical plot;

identifying a region of the canonical plot that corresponds to a classification of a group criterion; and

determining the group criterion classification of the object based on the location of the extracted data in the canonical plot relative to the identified region.

14. The method of claim 13, wherein receiving a selection of variables comprises:

designating the plurality of sets of known object data as a Y variables in a validated discriminant analysis;

receiving a selection of a group criterion as an X variable in a validated discriminant analysis; and

receiving a selection of other validated discriminant analysis variables.

15. The method of claim 14, wherein other validated discriminant analysis variables comprise at least one of validation type, number of folds, and stepwise selection.

16. The method of claim 13, wherein thee group criterion comprises one of sex, age-class or species.

17. The method of claim 13, wherein the output data representative of the determined classification of the object comprises an indication of the group criterion classification of the object.

18. A method of classifying an object comprising:

receiving a set of images of an indication of the object;

extracting image data from each image comprising the set of images of the indication of the object;

comparing the image data to a set of comparison data, wherein the set of comparison data comprises a plurality of data sets, wherein each of the plurality of data sets is associated with a comparative object, wherein comparing the image data to a set of comparison data comprises;

generating, through discriminant analysis of the image data and the set of comparison data, an image set of canonical variates corresponding to the image data, a plurality of comparison sets of canonical variates, wherein each of the plurality of comparison sets of canonical variates corresponds to a comparative object, and an RCV set of canonical variates corresponding to a cumulative set of image data and comparison data;

generating a first centroid value corresponding to the image set of canonical variates, a plurality of comparison centroid values, wherein each of the plurality of comparison centroid values corresponds to a comparison set of canonical variates of the plurality of comparison sets of canonical variates, and an RCV centroid value corresponding to the RCV set of canonical variates;

plotting, in two-dimensional space, the first set of canonical variates as a first canonical centroid plot represented by a first ellipse having a center point at the first centroid value, one of the plurality of comparison sets of canonical variates as a second canonical centroid plot represented by a second ellipse having a center point at one of the plurality of comparison centroid values, and the RCV set of canonical variates as an RCV canonical centroid plot, represented by a third ellipse having a center point at the RCV centroid value; and

determining whether the first ellipse overlaps the second ellipse;

determining a classification of the object based on the comparison; and

outputting, for display, output data representative of the determined classification of the object.

19. The method of claim 18, wherein comparing the image data set to a set of comparison data further comprises:

determining a plurality of distance values wherein a distance value comprises the distance between the first centroid value and a comparison centroid value of the plurality of comparison centroid values, the first centroid value and the RCV centroid value, a comparison centroid value of the plurality of comparison centroid values and the RCV value, or a first comparison centroid value of the plurality of comparison centroid values and a second comparison centroid value of the plurality of comparison centroid values;

applying a clustering technique to the plurality of distance values; and

generating, responsive to applying the clustering technique, a cluster dendrogram.

20. The method of claim 18, wherein the output data representative of the determined classification of the object comprises a cluster dendrogram.