Face Recognition

Used pre-trained Haar Cascade Classifier for facial object detection in the captured image and used K-Nearest Neighbor algorithm for face recognition

Face Recognition

Libraries used

  • numpy: for arrays/matrices
  • cv2: computer vision library

Image Recording

  • Source camera location is provided and the camera object is prepared to capture images from the source camera
  • Image frames are captured using the camera
  • The image frames are converted from BGR to grayscale
  • Objects of different sizes are detected from the input image and returned as a list using Haar-CascadeClassifier
  • For each face object we have the corner coordinates (x, y) and the width and height of the object
    • Get the component from the captured frame
    • The component is transform into a data by resizing into a data by resizing
    • Store the face data after every 10 frames, till we get 20 entries
    • Render a rectangle around the face in the frame for vizualization
    • Display the image frame in a window
    • If user presses the ‘Esc’ key or the number of images hits 20, then recording is stopped
  • The source camera is turned off and all the image frame windows created are destroyed
  • The data is saved as a numpy matrix in an encoded format

Image Recognizing

  • Source camera location is provided and the camera object is prepared to capture images from the source camera
  • Saved image data files are loaded and train data set is created
  • Labels are marked accordingly and a dictionary is created to map labels to the corresonding text string
  • A Function to calculate Euclidean distance is defined
  • K Nearest Neighbors algorithm to detect label is defined
  • Image frames are captured using the camera
  • The image frames are converted from BGR to grayscale
  • Objects of different sizes are detected from the input image and returned as a list using Haar-CascadeClassifier
  • For each face object we have the corner coordinates (x, y) and the width and height of the object
    • Get the component from the captured frame
    • The component is transform into a data by resizing into a data by resizing
    • The label of the data is predicted using KNN
    • The detected label is mapped to its corresponding text which is renedered on the frame
    • A rectangle is also rendered around the face in the frame for vizualization
    • The image frame is displayed in a window
    • If user presses the ‘Esc’ key, then recording is stopped
  • The source camera is turned off and all the image frame windows created are destroyed