Skip to content

Detect, recognize and verify faces using hybrid features: “deep” features from VGG-net + HoG + LBP. Hybrid Features help increase recognition significantly

Notifications You must be signed in to change notification settings

saravanabalagi/face_id

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

62 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Face ID

Result

What is it and how does this work?

Detect, recognize and verify faces using hybrid features: “deep” features from VGG-net + HoG + LBP. Uses VGG Face matconvnet for deep feature extraction along with features obtained from traditional methods HoG and LBP. Hybrid Features help increase accuracy of recognition to a significant level.

Detection

Can detect most faces. (Warning may have bias towards the data we trained it with)

Recognition

Can recognize the following faces:

  1. Abdullah Gul
  2. Amelie Mauresmo
  3. Andy Roddick
  4. Angelina Jolie
  5. Atal Bihari
  6. Bill Gates
  7. Bill Simon
  8. Britney Spears
  9. Carlos Menem
  10. Carlos Moya
  11. David Nalbandian
  12. Dick Cheney
  13. Dominique de
  14. Eduardo Duhalde
  15. Fidel Castro
  16. George Robertson
  17. Halle Berry
  18. Hamid Karzai
  19. Hillary Clinton
  20. Hu Jintao
  21. Igor Ivanov
  22. James Blake
  23. Jean Charest
  24. Jennifer Aniston
  25. Jennifer Lopez
  26. Jeremy Greenstock
  27. Jiang Zemin
  28. John Bolton
  29. John Howard
  30. John Kerry
  31. John Snow
  32. Joschka Fischer
  33. Jose Maria
  34. Julianne Moore
  35. Julie Gerberding

Trained with approximately 10-15 cropped face images of size 64x64 for each person.

Verification

Can verify if it is the same person given two images.

Trained with 1800 records each containing two faces and ground truth value of whether they were same or not.

Can it recognize and/or verify my face?

Yes...! All that's required is a few images of your face. Train the model with the same and Voila! More images = More Accurate verification

Marker Help Information

This mini-project was submitted for Image and Vision Computing Assignment

Files for marking

Assignment_1.m:
	Part 1: Face Detection

Assignment_2.m: 
	Part 2: Face Recognition and Verification

bonus.m:
	Bonus marks: Linked detector and recogniser, indentifies multiple individuals (drawn from validation dataset which the model has not seen before) correctly.

Dev Files

We split Assignment_2.m into two files for developmental convenience

Assignment_2_Part1.m: 
	Face Recognition

Assignment_2_Part2.m: 
	Face Verification

VGG Face model Location: ./library/matconvnet/data/models/vgg-face.mat

Demo Files

We made separate files for demo which loads the trained models

Assignment_1_eval.m: 
	Face Detection
	Loads model from ./face_detector.mat

Assignment_2_Part1_eval.m: 
	Face Recognition
	Loads model from ./models/fr_model.mat

Assignment_2_Part2_eval.m: 
	Face Verification
	Loads models from ./models/fv_model.mat