• 

  We also incorporated an open database of existing mugshots to detect crime recidivism

• 

  This is our homepage

• 

  After a fileupload, we search the facebook graph and profile pictures and return a sorted list of the profiles that match

• 

  An example showing Microsoft's Face API in action.

Social Recognition

A suspect identifying tool for law enforcement.

Identification through facial recognition technologies is not a new technology. When a crime occurs law enforcement often have tools at their disposal which allow officers to search through an existing database of mugshots. However this leads to a problem if the suspect has no criminal history.

Introducing ViridEssense, a play on the word Viridescent, or to become green. ViridEssence is an application which allows officers to identify suspects by comparing photos against a large database of social media photos in networks to include Facebook and Instagram.

How It Works

Once a day, we pull profile pictures from Facebook in North Carolina.

When a law enforcement official choses a file to upload, we send the image to Microsoft's Face API to extract critical demographic information for the officer. Information provided includes age, and gender. This approach offers an unbiased insight about the selected individual.

A photo is then uploaded to our website which contains an unidentified suspect. We then use dlib's state-of-the-art face recognition build in with deep learning to extract image features. Our model has an accuracy of 99.38% on the Labeled Faces in the Wild benchmark. Using this encoding, we compare it against the pre-computed encodings of profile images which we scraped from Facebook, Instagram, and other Social media sites.

Using surveillance photos from crime scenes at the Boston Marathon bombing, we successfully identified Dzhokhar Tsarnaev

With our pre-processing functionality which is run once a day, we are able to detect faces quickly and efficiently, and make the public safer by doing so. Our demo involves using public figures and mutual friends as test cases.

Built With

• computer-vision
• css
• html
• javascript
• neural-networks
• php
• python