z1311 / Image-Manipulation-Detection

Classifies a given image as authentic or tampered by doing two levels of analysis. Implemented using PyTorch.
21 stars 4 forks source link
computer-vision cv deep-learning error-level-analysis fake-image-detection fake-images metadata neural-networks python pytorch

Image-Manipulation-Detection

Classifies a given image as authentic or tampered by doing two levels of analysis.
Level 1 - Metadata analysis, to find any software signatures in the metadata of the image.
Level 2 - Feature Engineering(Error Level Analysis) and CNNs for classification. Error Level Analysis(ELA) is a compression method for finding the region which is tampared. This output is given as input to the CNN for classification.
The model is trained and validated on CASIA dataset. It has JPEG images of copy-move and spliced tampared images.

How To Run

Flags

-p or --path: Image pathname (required)

$ python main.py -p pathname