EpsilonPredictorDBSCAN

EpsilonPredictorDBSCAN.py

from cv2 import resize, INTER_LINEAR
from multiprocessing import Pool

def compare(args):
  img, img2 = args
  img = (img - img.mean()) / img.std()
  img2 = (img2 - img2.mean()) / img2.std()
  return np.mean(np.abs(img - img2))

class EpsilonPredictor:

  def __init__(self, features):
    self.features = [resize(img, (224, 224), INTER_LINEAR) for img in features]
    self.distances = np.zeros((len(features), len(features)))
  
  def measure(self, pool):
      for i, img in enumerate(self.features):
        all_imgs = [(img, f) for f in self.features]
        dists = pool.map(compare, all_imgs)
        self.distances[i, :] = dists
  
  def plot_result(self):
    plt.hist(self.distances.flatten(), bins=50)
    plt.title('Histogram of distance matrix')