musimab / ID_CardRecognition

Sift based face recognition

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

query images

rotated image

sift based recognition and warping

dlib face detection for both train and query image

matched points

matched ID cards

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

matched regions

Detected key points and feature comparision from cropped faces

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