i was trying to make as an exercise a model with scikit-learn to recognize images and return a box that surrounds the face. I got an SVC to work, but when trying to make a sliding window (my first time doing it) as result i got the image with multiple boxes that cover pretty much anything except the face, and in non-face pictures they get labeled as “face” inside the sliding window because like 3-4 window think they have a face in it, but when i test the model with the same pictures all works good.

I’m not sure if it’s a problem with the sliding window since i never used one and couldn’t find a source that explains how to do it properly, but i tried:

• Doing a loop for the lenghts of the images (x and y)
• Getting the window with custom size and custom step and resizing it as the model wanted, flatten and reshapen them.
• Then use the model to predict if the window is a face and get the rectangle.

def features_extr(window):
  window = resize(window, (50,50))
  window = window.flatten()
  window = window.reshape(1,-1)
  return window

def sliding_window(img,step,window_size,model,label):
  img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
  img_gray = cv2.equalizeHist(img_gray)

  for y in tqdm(range(0, img_gray.shape[0] - window_size[1] + 1, step)):
    for x in range(0, img_gray.shape[1] - window_size[0] + 1, step):
      window = img_gray[y:y + window_size[1], x:x + window_size[0]]

      features = features_extr(window)
      predict = model.predict(features)[0]

      if predict == label:
        cv2.rectangle(img, (x, y), (x + window_size[0], y + window_size[1]), (0, 255, 0), 2)
      
  plt.imshow(img)

sliding_window(img, 100, (100,100), gs, "Faces")