This is a repository for Hackrice8

This project uses DeepFace to do the one-shot face recognition, which can match two pictures and judge if they are the same person. In folder python, vgg_face.py, it can preprocess the input images and do verification.

• Make Sense of Data track uses some application of data science or machine learning
• Indeed Challenge innovative projects and solutions that tackle societal problems

Family members cannot get in contact with each other after natural disasters Disaster victims might be hospitalized without identification reconnect. first responders upload images and general information about disaster victims family members search for their loved ones by uploading photos facial-recognition matches the uploaded photos return top 10 possible matches, user selects correct profile

python=3.6.6

Matplotlib=2.2.2

tensorflow=1.7.0

keras=2.1.5

Cython=0.28.5

Download pre-trained model here

• Step 1: Transform images to 2622 dimensional vectors
• Step 2: Deep Learning Training and Inference
• Step 3: Vector similarity:

• Cosine distance

• Step 4: Verify Matches: Match = Distance less than threshold
• Step 5: Results: Top 10 most similar profiles

• Step 1: Upload Image
• Step 2: Run the facial matching algorithm back-end
• Step 3: Return top-10 most similar people
• Step 4: Fetch information and give feed back

• Areas for improvement more accurate facial recognition although our system deploys near state-of-the-art algorithmic architecture, we recognize that rapid advancements in technology may produce better infrastructure in the near future

• Future implementations ability to be notified of new matches store user-input images in a separate database ⇒ run matching algorithm for newly uploaded profiles ⇒ email/phone alert when new matches are found