Estimating PRNU on Gaussian noise images #21

New Issue

Closed

opened 2024-03-25 18:01:00 +01:00 by Benjamin_Loison · 6 comments

Benjamin_Loison commented 2024-03-25 18:01:00 +01:00

Owner

Copy Link

Copy Source

Should obtain the PRNU by averaging Gaussian noise images by definition.

Let us first try to proceed this way.

As PIL does not seem to allow us to easily see multiple images at the same time to compare them, let us use matplotlib to do so.

import matplotlib.pyplot as plt
import numpy as np

# Some example data to display
x = np.linspace(0, 2 * np.pi, 400)
y = np.sin(x ** 2)

fig, (ax1, ax2) = plt.subplots(1, 2)
fig.suptitle('Horizontally stacked subplots')
ax1.plot(x, y)
ax2.plot(x, -y)

plt.show()

Source: https://matplotlib.org/3.8.0/gallery/subplots_axes_and_figures/subplots_demo.html

Should compute RMS to see the estimated PRNU matches the actual one.

Related to #19.

Should obtain the PRNU by averaging Gaussian noise images by definition. Let us first try to proceed this way. As PIL does not seem to allow us to easily see multiple images at the same time to compare them, let us use matplotlib to do so. ```py import matplotlib.pyplot as plt import numpy as np # Some example data to display x = np.linspace(0, 2 * np.pi, 400) y = np.sin(x ** 2) fig, (ax1, ax2) = plt.subplots(1, 2) fig.suptitle('Horizontally stacked subplots') ax1.plot(x, y) ax2.plot(x, -y) plt.show() ``` Source: https://matplotlib.org/3.8.0/gallery/subplots_axes_and_figures/subplots_demo.html Should compute RMS to see the estimated PRNU matches the actual one. Related to #19.

Benjamin_Loison pinned this 2024-03-25 18:01:15 +01:00

Benjamin_Loison commented 2024-03-25 20:46:39 +01:00

Author

Owner

Copy Link

Copy Source

Showcase an example and now make a curve to make it clearer. Note that if want a maximum precision curve, may have to pay attention to complexity.

Showcase an example and now make a curve to make it clearer. Note that if want a maximum precision curve, may have to pay attention to complexity.

Benjamin_Loison commented 2024-03-26 01:24:31 +01:00

Author

Owner

Copy Link

Copy Source

Could render intermediary image differences, as it is easy to see an almost black uniform image but it is hard to compare 2 random images. However, this assumes that the color scale for each image is identical, which is probably not the case currently.

Could render intermediary image differences, as it is easy to see an almost black uniform image but it is hard to compare 2 random images. However, this assumes that the color scale for each image is identical, which is probably not the case currently.

Benjamin_Loison commented 2024-03-26 01:39:06 +01:00

Author

Owner

Copy Link

Copy Source

plt.title('Actual PRNU')
plt.imshow(prnus[0])
plt.show()

image = np.array(imagesWithPrnu[0][:1]).mean(axis = 0)
plt.title(f'First image with PRNU\ni.e. estimated PRNU\nRMS with actual PRNU = {round(rmsDiffNumpy(image, prnus[0]), 4)}')
plt.imshow(image)
plt.savefig('first_image_with_prnu.svg')

Could make the PRNU actually show PRNU:

Maybe would have to consider the transparency and not anti-aliasing.

See prnu_written_as_such branch.

```py plt.title('Actual PRNU') plt.imshow(prnus[0]) plt.show() ``` ```py image = np.array(imagesWithPrnu[0][:1]).mean(axis = 0) plt.title(f'First image with PRNU\ni.e. estimated PRNU\nRMS with actual PRNU = {round(rmsDiffNumpy(image, prnus[0]), 4)}') plt.imshow(image) plt.savefig('first_image_with_prnu.svg') ```          Could make the PRNU actually show `PRNU`:   Maybe would have to consider the transparency and not anti-aliasing. See [prnu_written_as_such](https://gitea.lemnoslife.com/Benjamin_Loison/Robust_image_source_identification_on_modern_smartphones/src/branch/prnu_written_as_such) branch.         

single_prnu_estimation_with_images_being_gaussian_noise.svg

217 KiB

single_prnu_estimation_with_images_being_gaussian_noise_tight_layout.svg

52 KiB

actual_prnu.svg

41 KiB

first_image_with_prnu.svg

51 KiB

mean_of_first_1_000_images_with_prnu.svg

52 KiB

mean_of_first_10_000_images_with_prnu.svg

51 KiB

mean_of_first_10_images_with_prnu.svg

52 KiB

mean_of_first_100_images_with_prnu.svg

53 KiB

first_image_without_prnu.svg

46 KiB

image.png

1.2 KiB

image.png

432 B

single_prnu_estimation_with_images_being_gaussian_noise_tight_layout.svg

52 KiB

single_prnu_estimation_with_images_being_gaussian_noise.svg

188 KiB

mean_of_first_10_images_with_prnu.svg

50 KiB

mean_of_first_100_images_with_prnu.svg

51 KiB

mean_of_first_1_000_images_with_prnu.svg

52 KiB

mean_of_first_10_000_images_with_prnu.svg

50 KiB

first_image_with_prnu.svg

51 KiB

actual_prnu.svg

20 KiB

first_image_without_prnu.svg

46 KiB

Benjamin_Loison referenced this issue from a commit 2024-03-26 01:55:10 +01:00

#21: Incorrect mean, as it is a RMS mean, not a mean of images

Benjamin_Loison referenced this issue from a commit 2024-03-26 01:59:23 +01:00

#21: Make a RMS curve depending on the number of images considered for the mean

Benjamin_Loison commented 2024-03-26 02:03:00 +01:00

Author

Owner

Copy Link

Copy Source

plt.xscale('log')

   ```py plt.xscale('log') ```

rms_between_actual_prnu_and_the_mean_of_the_first_n_images_with_prnu.svg

34 KiB

rms_between_actual_prnu_and_the_mean_of_the_first_n_images_with_prnu_log_x_and_y.svg

60 KiB

rms_between_actual_prnu_and_the_mean_of_the_first_n_images_with_prnu_log_x.svg

43 KiB

Benjamin_Loison added the epichigh priority labels 2024-03-26 02:13:52 +01:00

Benjamin_Loison referenced this issue from a commit 2024-03-26 03:15:36 +01:00

#21: Write PRNU as such to make figures clearer

Benjamin_Loison commented 2024-03-26 11:58:20 +01:00

Author

Owner

Copy Link

Copy Source

What is the next step?

Get closer to the real case with the 16 images of https://mcolom.perso.math.cnrs.fr/pages/no_noise_images/.
It is assumed that by their processing, these images do not sound. Thus we add a UNPR and potentially Gaussian noise and we hope that by defusing we will have on the one hand a noiseless image and on the other hand the UNPR, the Gaussian noise and an image residue. As a result, the UNPR will be retained but the Gaussian noise will cancel and it is hoped that the residual image by the number and diversity of scenes on the images will also cancel.
Note that the depressor, depending on the generation of the NRP, can interpret it as stage and therefore not be isolated as desired. To avoid this one can generate a UNPR more likely to be detected as noise by generating it as a Gaussian noise.

From Minutes Google Docs.

Translated with https://libretranslate.com.

What is the next step? > Get closer to the real case with the 16 images of https://mcolom.perso.math.cnrs.fr/pages/no_noise_images/. It is assumed that by their processing, these images do not sound. Thus we add a UNPR and potentially Gaussian noise and we hope that by defusing we will have on the one hand a noiseless image and on the other hand the UNPR, the Gaussian noise and an image residue. As a result, the UNPR will be retained but the Gaussian noise will cancel and it is hoped that the residual image by the number and diversity of scenes on the images will also cancel. Note that the depressor, depending on the generation of the NRP, can interpret it as stage and therefore not be isolated as desired. To avoid this one can generate a UNPR more likely to be detected as noise by generating it as a Gaussian noise. From `Minutes` Google Docs. Translated with https://libretranslate.com.

Benjamin_Loison commented 2024-03-28 15:59:57 +01:00

Author

Owner

Copy Link

Copy Source

As a result continue working at #25.

As a result continue working at #25.

Benjamin_Loison closed this issue 2024-03-28 16:00:18 +01:00

Benjamin_Loison unpinned this 2024-03-28 16:00:23 +01:00

Benjamin_Loison pinned this 2024-03-28 16:00:35 +01:00

Benjamin_Loison referenced this issue 2024-03-28 21:59:25 +01:00

Generate arbitrary shape `PRNU` shown as such #27

Benjamin_Loison unpinned this 2024-03-30 00:08:13 +01:00

Sign in to join this conversation.

No Branch/Tag Specified

Branches Tags

master

prnu_written_as_such

evaluation

Labels

Clear labels

bug

Something is not working

Context-Adaptive Interpolator

duplicate

This issue or pull request already exists

enhancement

New feature

epic

A task that is going to take more than a day to complete.

help wanted

Need some help

high priority

invalid

Something is wrong

left for future work

low priority

medium

A task that should take less than a day to complete.

medium priority

meta

question

More information is needed

quick

A task that should take less than two hours to complete.

wontfix

This won't be fixed

No labels epic high priority

Milestone

No items

No Milestone

Projects

Clear projects

No projects

Assignees

Clear assignees

Benjamin_Loison (Benjamin Loison)

No Assignees

1 Participants

Notifications

Subscribe

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Benjamin_Loison/Robust_image_source_identification_on_modern_smartphones#21