Skip to content

Latest commit

 

History

History
51 lines (46 loc) · 3.78 KB

README.md

File metadata and controls

51 lines (46 loc) · 3.78 KB

Image Similarity in Percentage %

Siamese network to compare image similarity in percentage - based on Keras deep learning model (VGG16, ResNet50) & cosine similarity, euclidean similarity

Accuracy

The cosine similarity and euclidean similarity are shown in the table.

image1 image2 cosine similarity (VGG16) euclidean similarity (VGG16) cosine similarity (ResNet50) euclidean similarity (ResNet50)
84.51% 0.01326 91.28% 0.05116
--- --- --- --- --- ---
63.95% 0.00980 54.98% 0.02871
--- --- --- --- --- ---
100.00% 1.0 100.00% 1.0
--- --- --- --- --- ---
63.66% 0.01222 78.96% 0.03771
--- --- --- --- --- ---
51.74% 0.01105 51.18% 0.02189
--- --- --- --- --- ---
23.80% 0.00907 30.91% 0.01755
--- --- --- --- --- ---
42.68% 0.01361 49.00% 0.02593
--- --- --- --- --- ---
69.20% 0.01478 70.07% 0.02849
--- --- --- --- --- ---
77.01% 0.02064 82.51% 0.04565
--- --- --- --- --- ---
Original Image Cropped Image 93.75% 0.03695 95.31% 0.07801
--- --- --- --- --- ---
Original Image Adversarial Image 74.47% 0.01384 90.14% 0.06188
--- --- --- --- --- ---
Original Image Adversarial Image 79.60% 0.01324 91.45% 0.04503
--- --- --- --- --- ---
Original Image Screenshot Image 97.95% 0.06415 98.69% 0.13120

License

Citations

@inproceedings{Panagiotis2021,
  author = {Panagiotis Kasnesis, Ryan Heartfield, Xing Liang, Lazaros Toumanidis, Georgia Sakellari, Charalampos Patrikakis, George Loukas},
  booktitle = {Journal of Applied Soft Computing},
  title = {Transformer-based identification of stochastic information cascades in social networks using text and image similarity},
  year = {2021}
}