Skip to content

musimab/ID_CardRecognition

Repository files navigation

Sift based face recognition from ID cards

Implementation

  1. Read train and query images
  2. Extract sift features
  3. Calculate boundries using homograpy transform
  4. find orintation of ID card
  5. Rotate ID card and warp
  6. Run dlib face detector
  7. Crop face regions
  8. Extract sift features from cropped faces
  9. Compare faces with Lowe's ratio

input image

testcard

query images

tc_ID tc_ID_rot

rotated image

rotated_img

sift based recognition and warping

warped_img

dlib face detection for both train and query image

face_crop_target

cropped_fc

matched points

Figure_1

matched ID cards

matched_points_ID

Detection Results

  1. Total good matches: 134
  2. Rotation Angle: -43 degrees
  3. Number of Faces: 1 (input image)
  4. Number of Faces: 1 (output image)
  5. Total good matches: 23
  6. Matches are found - 10/10
  7. Faces are similar

input image

Lets change our input image test3

matched regions

matched_ID_

Detected key points and feature comparision from cropped faces

face_matched

  1. Total good matches: 96
  2. Rotation Angle: -43 degrees
  3. Number of Faces: 1 (input image)
  4. Number of Faces: 1 (query image)
  5. Total good matches: 0
  6. Matches are found - 1/15
  7. Faces are not similar