Comparison between two images

[riclab]

Hi, is it possible to make a comparison between two images getting a percentage?

Thanks.

[aleju]

There isn't a function for that, but it shouldn't be hard to implement. Assuming you have trained the model and the weights were saved under filepath weights_fp and your images are at filepaths img1_fp and img2_fp, then the process should roughly be (not tested):

from scipy import ndimage
from skimage import color
from train import create_model

# create the model
model = create_model() # that function is in train.py

# load trained weights
model.load_weights(weights_fp)

# load images
img1 = ndimage.imread(img1_fp, mode="RGB")
img2 = ndimage.imread(img2_fp, mode="RGB")

# resize images
img1_rs = misc.imresize(img1, (32, 32))
img2_rs = misc.imresize(img2, (32, 32))

# convert images to grayscale images with shape (height, width, 1) in value range 0 to 1.0
img1_gray = color.rgb2gray(img1_rs)[:, :, np.newaxis]
img2_gray = color.rgb2gray(img2_rs)[:, :, np.newaxis]

# convert images to batches
img1_batch = np.array([img1_gray])
img2_batch = np.array([img2_gray])

# predict their similarity
preds = model.predict([img1_batch, img2_batch])

# print similarity
print("Similarity: %.2f" % (preds[0],))

[riclab]

Thank you, Alexander!

[shubhpy]

@riclab Did you run the script?
It is throwing error "Can't open attribute (can't locate attribute: 'layer_names')" on model.load_weights step.