Summer Semester 2020 Project
Face Morphing Attacks (FMA) pose a serious threat and other concerns to current travel regulations security. In this piece of work, we demonstrate how a live image of the user face acquired at uncontrolled environment, can be used to restore the de-morphed image from the morphed image stored in the travel document. A series of steps carried out on a data set gives the benchmark for further analysis and results.
The application follows a generic workflow as described below.For developing the Android application to detect potentially morphed passport images, we started by portraying a layoutfor the application followed by implementing the features one by one.
The steps are as follows:
- Read the travel document data (image in our case) usingNFC
- Capture a live image with device camera
- Perform a face match
- If necessary perform De-morphing process
- Visualization and Evaluation of Results
For Reading face images from a document using NFC we are going to use jmrtd (https://jmrtd.org/) For taking a live-image with the camera we used CameraX module using a fixed ratio with guidelines to make sure every face is in the same position For matching we are using MobileFaceNet architecture with help of tensorflow lite and MTCNN for detection and place landmark on the face. Please refer to the report file for detailed information.
- First Scan the image from ePassport by clicking "NFC SCAN" Button. Enter required details and click on "SAVE IMAGE" button on right top corner.
- Then Live Camera Image can be taken by "CAPTURE" button.
- After saving both images, "Match" will be enabled and similarity between two faces can be measured after detecting the face ("DETECT") from images.
- Upon successful matching and similarity score more than predefined threshold, DEMORPH oprtion will be enabled.
- By pressing "DEMORPH" it will send the data to server and result with de-morphed image will be apear. Note :For successful demorph process, the smartphone needs to be connected through university network.