METHODS FOR IDENTIFYING IMAGING DEVICES AND CLASSIFYING IMAGES ACQUIRED BY UNKNOWN IMAGING DEVICES

• US 20120087589A1
• Filed: 02/15/2010
• Published: 04/12/2012
• Est. Priority Date: 02/13/2009
• Status: Active Grant

• Current Assignee: Forensic Pathways Limited
• Original Assignee: Forensic Pathways Limited
• Inventors: Chang-Tsun, Li, Leary, Richard
• Granted Patent: US 8,565,529 B2
• Time in Patent Office: Days
• US Class Current: 382/190
• CPC Class Codes: G06F 18/23211 with adaptive number of clusters
• CPC Class Codes: G06V 10/30 Noise filtering
• CPC Class Codes: G06V 10/763 Non-hierarchical techniques, e.g. based on statistics of modelling distributions

Abstract

A method of classifying an image taken by an image capture device, the method comprising the steps of: extracting an initial Sensor Noise Pattern (SNP) for the image; enhancing the initial SNP to create an enhanced SNP by applying a correcting model, wherein the correcting model scales the initial SNP by a factor inversely proportional to the signal intensity of the initial SNP; determining a similarity measure between the enhanced SNP for said image with one or more previously calculated enhanced SNPs for one or more different images; and classifying the image in a group of one or more images with similar or identical SNPs based on the determined similarity measure.

Claims

1. A method of classifying an image taken by an image capture device, the method comprising the steps of:

extracting an initial Sensor Noise Pattern (SNP) for the image;

enhancing the initial SNP to create an enhanced SNP by applying a correcting model, wherein the correcting model scales the initial SNP by a factor which decreases as the magnitude of signal intensity of the initial SNP increases;

determining a similarity measure between the enhanced SNP for said image with one or more previously calculated enhanced SNPs for one or more different images; and

classifying the image in a group of one or more images with similar or identical SNPs based on the determined similarity measure.

2. A method of classifying a plurality images taken by one or more known or unknown image capture devices, the method comprising the steps of:

for each image

classifying the image based on the method of claim 1;

identifying the image in a group of one or more images and identifying in a subset of images from the plurality of images; and

forming an image classifier based on the images which have been classified as having identical or similar SNPs,

classifying one or more of the remaining images that were not part of the initial subset by calculating a comparative measure between the remaining images and each cluster as identified in the image classifier, and determining if said remaining image belongs to an identified cluster based on the comparative measure.

3. The method of claim 2, wherein the identification of the clusters in the subset of images comprises the steps of, for each image of the subset:

calculating a similarity matrix to determine a comparative measure of the similarity of the enhanced SNPs for each image that forms the subset;

identifying an initial set of intra-cluster members and inter-cluster members of the subset of images by means of a k-means algorithm;

determining a reference similarity based on a measure of the difference between identified the intra and inter-clusters;

identifying a membership committee and for each image in the membership committee;

calculating a cost score based on their comparative measure and reference similarity, and updating the intra-cluster members based on the cost score, thereby identifying groups of images which have similar or identical enhanced SNPs.

4. The method of claim 3 further comprising:

assigning each image in the subset a unique class label; and

updating the class label for images which are identified as having similar or identical SNPs by the cost score with the same class label.

5. The method of claim 2 wherein images which have a similarity measure below a threshold are identified as not belonging to a predetermined cluster and are identified as being founding members of a new cluster.

6. The method of claim 1 where the initial extraction SNP, and subsequent enhancement and comparison are performed on a subsection of the input and comparison images.

7. The method of claim 1 wherein the initial SNP is identified with a wavelet based de-noising filter and the correcting model is one or more of the following functions:

8. The method of claim 7 wherein α

=7.

9. The method of claim 1 wherein the similarity measurement is based on a comparative measure of the input and reference images.

11. The method of claim 9 wherein the comparative measure is a vector having the form:

12. The method of claim 2 wherein the identification of groups of images with identical or similar SNPs occurs by a k-means clustering algorithm based on the comparative measure.

13. The method of claim 2 wherein the reference similarity is a measure of the distance between the centroids defined by the intra-cluster members and inter-cluster members and wherein the centroids are updated when the intra-cluster members are updated.

16. The method of claim 2 further comprising the steps of:

identifying images taken from a given source image capture device by identifying images which have a similar or identical SNP to that of the source image capture device;

creating a reference SNP for the source image capture device;

determining a similarity measure between the reference SNP for the source image capture device and the identified clusters; and

identifying a cluster of images, if any, which originate from said source image capture device based on the similarity measure.

17. The method of claim 2 further comprising the steps of:

identifying images that show similar or identical characteristics to create a reduced subset of images from the plurality of images; and

performing the analysis on the reduced subset.

18. The method of claim 16 where the characteristics are based on one or more of the following:

make of camera, model of camera, image size, time that the image was taken, and date that the image was taken as identified by some form of metadata associated with the image, and

keywords or tags associated with the image, said keywords or tags being indicative of the content of the image.

20. The method of claim 17 wherein the keywords or tags are assigned to a image by a user after inspection of the image.

21. A method of identifying a plurality of images as originating from the same image capture device the method comprising:

classifying the image according to the method of claim 1; and

further comprising the step of identifying a classified group of images as originating from the same image capture device.

22. The method of claim 17 wherein the similarity measure between the image i and the comparison image j is determined by

25. A computer program product comprising computer readable instructions encoded thereon to enable the steps of method claim 1.

10. (canceled)

14. (canceled)

15. (canceled)

19. (canceled)

23. An apparatus for classifying an image taken by an imaging capturing device, the apparatus comprising:

a processor enabled to extract an initial Sensor Noise Pattern (SNP) from the image;

an enhancer enabled to enhancer the initial SNP to create an enhanced SNP by applying a correcting model, wherein the correcting model scales the initial SNP by a factor which decreases as the signal intensity of the initial SNP increases;

a similarity measurer to determine the similarity between the enhanced SNP for said image with one or more previously calculated enhanced SNPs for one or more different images; and

classifier to group of one or more images with similar or identical SNPs based on the determined similarity measure.

24. (canceled)