METHOD AND APPARATUS FOR REMOVING HIDDEN DATA BASED ON AUTOREGRESSIVE GENERATIVE MODEL

• US 20200134774A1
• Filed: 01/10/2019
• Published: 04/30/2020
• Est. Priority Date: 10/24/2018
• Status: Active Grant

• Current Assignee: Seoul National University R&DB Foundation (Seoul National University)
• Original Assignee: Seoul National University R&DB Foundation (Seoul National University)
• Inventors: YOON, Sungroh, BAE, Ho, JUNG, Dahuin
• Granted Patent: US 10,909,649 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 3/048 Activation functions
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: G06T 1/0021 Image watermarking
• CPC Class Codes: G06T 1/0028 Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
• CPC Class Codes: G06T 2201/0051 Embedding of the watermark in the spatial domain
• CPC Class Codes: G06T 2201/0065 Extraction of an embedded watermark; Reliable detection
• CPC Class Codes: G06T 2201/0201 whereby only tamper or origin are detected and no embedding takes place
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 5/00 Image enhancement or restoration
• CPC Class Codes: G06T 7/13 Edge detection

Abstract

Disclosed is a hidden data removal method based on an autoregressive generative model which is performed by a computer device. The hidden data removal method includes receiving a source image, randomly selecting a target pixel from the source image, and inputting the source image and an identifier of the target pixel to an autoregressive generative model and restoring the target pixel from the source image. The source image is an image in which steganography-based data is hidden, and the autoregressive generative model restores the target pixel on the basis of a pixel value distribution for pixels adjacent to the target pixel in the source image.

Claims

1. A method of removing hidden data based on an autoregressive generative model, the method being performed by a computer device, the method comprising:

receiving a source image;

randomly selecting a target pixel from the source image; and

inputting the source image and an identifier of the target pixel to the autoregressive generative model and restoring the target pixel,

wherein the source image is an image in which steganography-based data is hidden, and the autoregressive generative model restores the target pixel on the basis of a pixel value distribution for pixels adjacent to the target pixel in the source image.

2. The method of claim 1, wherein the computer device selects a plurality of target pixels from the source image.

3. The method of claim 1, further comprising detecting an edge of the source image,

wherein the computer device selects the target pixel from the edge.

4. The method of claim 1, wherein the autoregressive generative model is PixelCNN or Gated PixelCNN.

5. The method of claim 1, wherein the autoregressive generative model determines a pixel value of the target pixel such that at least one of a peak signal-to-noise ratio (PSNR) and a structure similarity index (SSIM) is maintained at a reference value or greater on the basis of a likelihood of the pixel value distribution.

6. The method of claim 1, wherein an input characteristic map and an output characteristic map have the same size, and the autoregressive generative model performs convolution using only some input characteristics.

7. The method of claim 1, wherein the source image has a standardized format, and the autoregressive generative model is prepared by pre-learning an image distribution of the source image.

8. The method of claim 1, wherein the computer device restores the target pixel on the basis of the adjacent pixels, which are included in a window having a center at which the target pixel is located.

16. A computer-readable recording medium having a computer program recorded thereon to execute the method of claim 1.

9. An apparatus for removing hidden data based on an autoregressive generative model, the apparatus comprising:

an input device configured to receive a source image in which steganography-based data is hidden;

a storage device configured to store an autoregressive generative model for restoring a specific pixel on the basis of a pixel value distribution for pixels adjacent to the specific pixel in an image; and

a computation device configured to select a plurality of target pixels from the source image, input the source image and an identifier of the target pixel to the autoregressive generative model, and restore the target pixel.

10. The apparatus of claim 9, wherein the autoregressive generative model is PixelCNN or Gated PixelCNN.

11. The apparatus of claim 9, wherein the autoregressive generative model determines a pixel value of the target pixel such that at least one of a peak signal-to-noise ratio (PSNR) and a structure similarity index (SSIM) is maintained at a reference value or greater on the basis of a likelihood of the pixel value distribution.

12. The apparatus of claim 9, wherein an input characteristic map and an output characteristic map have the same size, and the autoregressive generative model performs convolution using only some input characteristics.

13. The apparatus of claim 9, wherein the source image has a standardized format, and the autoregressive generative model is prepared by pre-learning an image distribution of the source image.

14. The apparatus of claim 9, wherein the computation device restores the target pixel on the basis of the adjacent pixels, which are included in a window having a center at which the target pixel is located.

15. The apparatus of claim 9,

wherein the storage device further stores an edge detection program for detecting an edge of an image, and

wherein the computation device detects an edge of the source image using the edge detection program and then selects the target pixel from the edge.