Determining image forensics using gradient statistics at edges

• US 10,586,152 B2
• Filed: 02/16/2017
• Issued: 03/10/2020
• Est. Priority Date: 02/16/2017
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Chen, Can, McCloskey, Scott
• Primary Examiner(s): Hu, Fred H
• Application Number:
  US15/435,066
• Publication Number:
  US 20180232605A1
• Time in Patent Office: 1,118 Days
• Field of Search: None
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06V 10/44 Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
• CPC Class Codes: G06V 10/50 by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
• CPC Class Codes: G06V 10/82 using neural networks

Abstract

A system assesses the integrity of a digital image by detecting an edge in the digital image and defining a patch of pixels encompassing the edge. The system then generates data relating to intensity and gradient magnitude for pixels in the patch, analyzes the data relating to intensity and gradient magnitude, and determines that the digital image has been forged or the digital image has not been forged based on the analysis of the data relating to intensity and gradient magnitude.

Claims

1. A process to assess integrity of a digital image comprising:

detecting an edge in the digital image;

defining a patch of pixels encompassing the edge;

generating data relating to intensity and gradient magnitude for pixels in the patch;

analyzing the data relating to intensity and gradient magnitude; and

determining that the digital image has been forged or the digital image has not been forged based on the analysis of the data relating to intensity and gradient magnitude;

wherein a forged image comprises an alteration of digital content of the digital image;

wherein the generating data relating to intensity and gradient magnitude for the pixels in the patch comprises generating a scatter plot of the intensity versus the gradient magnitude for the pixels in the patch; and

wherein the analyzing the data and the determining that the digital image has been forged or the digital image has not been forged comprises processing the data relating to pixel intensity and gradient magnitude via a machine learning model, and determining from output of the machine learning model that the digital image has been forged or the digital image has not been forged; and

comprising training the machine learning model by;

generating a two-dimensional histogram from the scatter plot, processing data from the two-dimensional histogram via a support vector machine (SVM), and determining from the SVM that the digital image has been forged or the digital image has not been forged;

gathering a plurality of two-dimensional histograms;

generating a first set of forged training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a symmetric arch, and wherein a gradient magnitude from low intensity to middle intensity and a gradient magnitude from middle intensity to high intensity are substantially the same;

generating a second set of authentic training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a skewed asymmetric arch, and wherein the gradient magnitude from low intensity to middle intensity and the gradient magnitude from middle intensity to high intensity are not substantially the same; and

training a forgery detection model using the first set of forged training data and the second set of authentic training data.

2. The process of claim 1, wherein the analyzing the data and the determining that the digital image has been forged or the digital image has not been forged comprises comparing the data to parametric models of optically blurred digital images, artificially blurred digital images, and artificially sharp digital images, and determining whether the digital image has been forged or has not been forged based on the comparing.

3. The process of claim 1, comprising generating a third set of forged training data by accumulating data from the two-dimensional histograms wherein the two dimensional histograms include an approximate right angle;

and training the forgery detection model using the first set of forged training data, the second set of authentic training data, and the third set of forged training data.

4. The process of claim 1, comprising processing the data from the two-dimensional histogram via a convolutional neural network (CNN), and determining from the CNN that the digital image has been forged or the digital image has not been forged.

5. The process of claim 4, comprising:

gathering a plurality of two-dimensional histograms;

generating a first set of forged training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a symmetric arch, and wherein a gradient magnitude from low intensity to middle intensity and a gradient magnitude from middle intensity to high intensity are substantially the same;

generating a second set of authentic training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a skewed asymmetric arch, and wherein the gradient magnitude from low intensity to middle intensity and the gradient magnitude from middle intensity to high intensity are not substantially the same; and

training a forgery detection model using the first set of forged training data and the second set of authentic training data.

6. The process of claim 5, comprising generating a third set of forged data by accumulating data from the two-dimensional histograms wherein the two dimensional histograms include an approximate right angle;

and training the forgery detection model using the first set of forged training data, the second set of authentic training data, and the third set of forged training data.

7. The process of claim 1, wherein the determining that the digital image has been forged or the digital image has not been forged comprises fusing the analysis for a plurality of patches, and determining whether the digital image has been forged or the digital image has not been forged based on the fusing.

8. The process of claim 1, comprising:

identifying a plurality of edges that are related;

generating a mask using the plurality of edges; and

identifying a location of a forged object in the digital image using the mask.

9. The process of claim 1, comprising training a forgery detection model by providing to the forgery detection model data relating to optically blurred edges from an authentic digital image, data relating to artificially blurred edges from a known forged image, and data relating to artificially sharp edges from a known forged image.

10. The process of claim 9, wherein the forgery detection model comprises one of more of a parametric fit operation, a support vector machine (SVM), and a convolutional neural network.

11. The process of claim 9, wherein the training comprises a multi-dimensional training scheme comprising comparing optically blurred edges to artificially blurred edges and artificially sharp edges, comparing artificially blurred edges to optically blurred edges and artificially sharp edges, and comparing artificially sharp edges to optically blurred edges and artificially blurred edges.

12. The process of claim 1, comprising processing the patch of pixels via a convolutional neural network (CNN), and determining from the CNN that the digital image has been forged or the digital image has not been forged.

13. The process of claim 1, comprising:

determining a first intensity on a first side of the edge;

determining a second intensity on a second side of the edge; and

predicting an amount of asymmetry expected in an authentic image based on the first intensity and the second intensity.

14. A system comprising:

a computer processor and a computer storage device configured for;

detecting an edge in the digital image;

defining a patch of pixels encompassing the edge;

generating data relating to intensity and gradient magnitude for pixels in the patch;

analyzing the data relating to intensity and gradient magnitude; and

determining that the digital image has been forged or the digital image has not been forged based on the analysis of the data relating to intensity and gradient magnitude;

wherein a forged image comprises an alteration of digital content of the digital image;

wherein the generating data relating to intensity and gradient magnitude for the pixels in the patch comprises generating a scatter plot of the intensity versus the gradient magnitude for the pixels in the patch; and

wherein the analyzing the data and the determining that the digital image has been forged or the digital image has not been forged comprises processing the data relating to pixel intensity and gradient magnitude via a machine learning model and determining from output of the machine learning model that the digital image has been forged or the digital image has not been forged; and

comprising training the machine learning model by;

generating a two-dimensional histogram from the scatter plot, processing data from the two-dimensional histogram via a support vector machine (SVM), and determining from the SVM that the digital image has been forged or the digital image has not been forged;

gathering a plurality of two-dimensional histograms;

generating a first set of forged training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a symmetric arch, and wherein a gradient magnitude from low intensity to middle intensity and a gradient magnitude from middle intensity to high intensity are substantially the same;

generating a second set of authentic training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a skewed asymmetric arch, and wherein the gradient magnitude from low intensity to middle intensity and the gradient magnitude from middle intensity to high intensity are not substantially the same; and

training a forgery detection model using the first set of forged training data and the second set of authentic training data.

15. A non-transitory computer readable medium comprising instructions that when executed by a processor execute a process comprising:

detecting an edge in the digital image;

defining a patch of pixels encompassing the edge;

generating data relating to intensity and gradient magnitude for pixels in the patch;

analyzing the data relating to intensity and gradient magnitude; and

determining that the digital image has been forged or the digital image has not been forged based on the analysis of the data relating to intensity and gradient magnitude;

wherein a forged image comprises an alteration of digital content of the digital image;

wherein the generating data relating to intensity and gradient magnitude for the pixels in the patch comprises generating a scatter plot of the intensity versus the gradient magnitude for the pixels in the patch; and

wherein the analyzing the data and the determining that the digital image has been forged or the digital image has not been forged comprises processing the data relating to pixel intensity and gradient magnitude via a machine learning model and determining from output of the machine learning model that the digital image has been forged or the digital image has not been forged; and

comprising training the machine learning model by;

generating a two-dimensional histogram from the scatter plot, processing data from the two-dimensional histogram via a support vector machine (SVM), and determining from the SVM that the digital image has been forged or the digital image has not been forged;

gathering a plurality of two-dimensional histograms;

generating a first set of forged training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a symmetric arch, and wherein a gradient magnitude from low intensity to middle intensity and a gradient magnitude from middle intensity to high intensity are substantially the same;

generating a second set of authentic training data by accumulating data from the two-dimensional histograms wherein the two-dimensional histograms include a skewed asymmetric arch, and wherein the gradient magnitude from low intensity to middle intensity and the gradient magnitude from middle intensity to high intensity are not substantially the same; and

training a forgery detection model using the first set of forged training data and the second set of authentic training data.