Benjamin_Loison/Robust_image_source_identification_on_modern_smartphones
1
0
Fork 0
Code Issues 69 Pull Requests Packages Projects Releases 1 Wiki Activity

Make single figure to compare multiple splitNXN values

Browse Source
This commit is contained in:
Benjamin Loison
2024-03-29 13:47:27 +01:00
parent 81852d5dc6
commit f0db61efe2
1 changed files with 62 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

95
datasets/noise_free_test_images/estimate_prnu.py
View File

@@ -25,47 +25,76 @@ datasetPath = 'no_noise_images'
# In addition to the range difference, note that the distribution in the first set of images was a Gaussian and here is very different and specific.
PRNU_FACTOR = 0.01
NOISE_FACTOR = 0.1
SPLIT_N_X_N = 1
IMAGE_SIZE_SHAPE = [dimension // SPLIT_N_X_N for dimension in (704, 469)]
np.random.seed(0)
#prnuNpArray = 255 * randomGaussianImage(scale = PRNU_FACTOR, size = IMAGE_SIZE_SHAPE)
prnuNpArray = getPrnuShownAsSuch(IMAGE_SIZE_SHAPE, 255) * PRNU_FACTOR
SPLIT_N_X_N_S = [1, 2, 4]
def isIn256Range(x):
return 0 <= x and x <= 255
# len(SPLIT_N_X_N_S)
fig, axes = plt.subplots(2, 4)
fig.suptitle('PRNU estimation with different number of images having Gaussian noise and Gaussian noised PRNU')
imagesPrnuEstimateNpArray = []
for splitNXNIndex, splitNXN in enumerate(SPLIT_N_X_N_S):
IMAGE_SIZE_SHAPE = [dimension // splitNXN for dimension in (704, 469)]
for imageName in os.listdir(datasetPath):
if imageName.endswith('.png'):
imagePath = f'{datasetPath}/{imageName}'
imageWithoutPrnuPil = Image.open(imagePath).convert('F')
imageWithoutPrnuNpArray = np.array(imageWithoutPrnuPil)
#prnuNpArray = 255 * randomGaussianImage(scale = PRNU_FACTOR, size = IMAGE_SIZE_SHAPE)
prnuNpArray = getPrnuShownAsSuch(IMAGE_SIZE_SHAPE, 255) * PRNU_FACTOR
m = IMAGE_SIZE_SHAPE[1]
n = IMAGE_SIZE_SHAPE[0]
def isIn256Range(x):
return 0 <= x and x <= 255
imageWithoutPrnuNpArrayTiles = [imageWithoutPrnuNpArray[x : x + m, y : y + n] for x in range(0, imageWithoutPrnuNpArray.shape[0], m) for y in range(0, imageWithoutPrnuNpArray.shape[1], n)]
for imageWithoutPrnuNpArrayTile in imageWithoutPrnuNpArrayTiles:
#print(imageWithoutPrnuNpArrayTile.shape, tuple(IMAGE_SIZE_SHAPE[::-1]))
#if imageWithoutPrnuNpArrayTile.shape != tuple(IMAGE_SIZE_SHAPE[::-1]):
# continue
imageNoise = randomGaussianImage(scale = 255 * NOISE_FACTOR, size = imageWithoutPrnuNpArrayTile.shape)
imageWithPrnuNpArray = imageWithoutPrnuNpArrayTile + prnuNpArray + imageNoise
#assert all([isIn256Range(extreme) for extreme in [imageWithPrnuNpArray.max(), imageWithPrnuNpArray.min()]]), 'Adding the PRNU resulted in out of 256 bounds image'
imageWithPrnuPil = toPilImage(imageWithPrnuNpArray)
#imagePrnuEstimatePil = contextAdaptiveInterpolator(imageWithPrnuPil.load(), imageWithPrnuPil)
#imagePrnuEstimateNpArray = np.array(imagePrnuEstimatePil)
imagePrnuEstimateNpArray = imageWithPrnuNpArray - denoise_tv_chambolle(imageWithPrnuNpArray, weight=0.2, channel_axis=-1)
imagesPrnuEstimateNpArray = []
isFirstImage = True
imagesPrnuEstimateNpArray += [imagePrnuEstimateNpArray]
for imageName in os.listdir(datasetPath):
if imageName.endswith('.png'):
imagePath = f'{datasetPath}/{imageName}'
imageWithoutPrnuPil = Image.open(imagePath).convert('F')
imageWithoutPrnuNpArray = np.array(imageWithoutPrnuPil)
cameraPrnuEstimateNpArray = np.array(imagesPrnuEstimateNpArray).mean(axis = 0)
rms = rmsDiffNumpy(cameraPrnuEstimateNpArray, prnuNpArray, True)
title = f'Camera PRNU estimate\nRMS with actual one: {rmsDiffNumpy(cameraPrnuEstimateNpArray, prnuNpArray):.4f} (normalized RMS: {rmsDiffNumpy(cameraPrnuEstimateNpArray, prnuNpArray, True):.4f})'
#showImageWithMatplotlib(cameraPrnuEstimateNpArray, title)
m = IMAGE_SIZE_SHAPE[1]
n = IMAGE_SIZE_SHAPE[0]
showImageWithMatplotlib(cameraPrnuEstimateNpArray, title, 'gray')
imageWithoutPrnuNpArrayTiles = [imageWithoutPrnuNpArray[x : x + m, y : y + n] for x in range(0, imageWithoutPrnuNpArray.shape[0], m) for y in range(0, imageWithoutPrnuNpArray.shape[1], n)]
for imageWithoutPrnuNpArrayTile in imageWithoutPrnuNpArrayTiles:
#print(imageWithoutPrnuNpArrayTile.shape, tuple(IMAGE_SIZE_SHAPE[::-1]))
#if imageWithoutPrnuNpArrayTile.shape != tuple(IMAGE_SIZE_SHAPE[::-1]):
# continue
imageNoise = randomGaussianImage(scale = 255 * NOISE_FACTOR, size = imageWithoutPrnuNpArrayTile.shape)
imageWithPrnuNpArray = imageWithoutPrnuNpArrayTile + prnuNpArray + imageNoise
if splitNXNIndex == 0 and isFirstImage:
axis = axes[0]
axis[0].set_title('First image without noise')
axis[0].imshow(imageWithoutPrnuNpArrayTile)
axis[1].set_title('Actual Gaussian noised (G.n.) PRNU')
axis[1].imshow(prnuNpArray)
axis[2].set_title('F. image with G.n.')
axis[2].imshow(imageWithoutPrnuNpArray + imageNoise)
axis[3].set_title('F. image with G.n. and PRNU')
axis[3].imshow(imageWithoutPrnuNpArray + prnuNpArray + imageNoise)
isFirstImage = False
#assert all([isIn256Range(extreme) for extreme in [imageWithPrnuNpArray.max(), imageWithPrnuNpArray.min()]]), 'Adding the PRNU resulted in out of 256 bounds image'
imageWithPrnuPil = toPilImage(imageWithPrnuNpArray)
#imagePrnuEstimatePil = contextAdaptiveInterpolator(imageWithPrnuPil.load(), imageWithPrnuPil)
#imagePrnuEstimateNpArray = np.array(imagePrnuEstimatePil)
imagePrnuEstimateNpArray = imageWithPrnuNpArray - denoise_tv_chambolle(imageWithPrnuNpArray, weight=0.2, channel_axis=-1)
imagesPrnuEstimateNpArray += [imagePrnuEstimateNpArray]
cameraPrnuEstimateNpArray = np.array(imagesPrnuEstimateNpArray).mean(axis = 0)
rms = rmsDiffNumpy(cameraPrnuEstimateNpArray, prnuNpArray, True)
title = f'RMS with actual PRNU: {rmsDiffNumpy(cameraPrnuEstimateNpArray, prnuNpArray):.4f}\n(normalized RMS: {rmsDiffNumpy(cameraPrnuEstimateNpArray, prnuNpArray, True):.4f})'
axis = axes[1]
axis[splitNXNIndex].set_title(f'Number of images: {len(imagesPrnuEstimateNpArray)}\n{title}')
axis[splitNXNIndex].imshow(cameraPrnuEstimateNpArray)
axes[1][3].axis('off')
plt.tight_layout()
plt.show()