jiwookseo/local_binary_patterns

Computer Vision Feature Detection, LBP by python

Local Binary Patterns

Computer vision 분야의 Feature detection 방식 중 하나인 Local Binary Patterns

Face recognizernition 등에 사용된다.

I. Concept

이미지를 3x3 pixel의 블록으로 쪼개어 블록의 중앙값을 기준으로 theresholding 하여 binary로 uint8로 변환하는 방식

II. Implementation

• LBP Code

  import matplotlib.pyplot as plt
  import numpy as np
  
  org_img = np.transpose(plt.imread('iu3.jpeg'), (2, 0, 1))
  r, c = org_img[0].shape
  # Image Import Part
  # RGB channel
  # org_img.shape = (3, r, c)
  
  pd_img = np.pad(org_img, ((0, 0), (1, 1), (1, 1)),\
                  'constant', constant_values=(0, 0))
  # Padding Part
  # pd_img.shape = (3, r+1, c+1)
  # for i in range(1, r):
  #	for j in range(1, c):
  #		pd_img[i][j] = org_img[i][j]
  
  lbp_img = np.zeros_like(org_img)
  # result.shape = (3, r, c)
  
  for i in range(3):
      ch_img = pd_img[i]
      # channal choice
      for j in range(r):
          for k in range(c):
              t = ch_img[j:j + 3, k:k + 3]
              # 3*3 array
              tt = t >= t[1, 1]
              # compare with center value
              lbp_img[i][j][k] = \
                  np.packbits((tt[1, 0], tt[2, 0], tt[2, 1], tt[2, 2],\
                               tt[1, 2], tt[0, 2], tt[0, 1], tt[0, 0]))[0]
              # Boolean Array 2 Unsigned Int8
              # Into result image
              
  for i in range(3):
      plt.subplot(2, 3, i + 1)
      plt.imshow(org_img[i], cmap=plt.get_cmap('gray'))
      plt.subplot(2, 3, 3 + i + 1)
      plt.imshow(lbp_img[i], cmap=plt.get_cmap('gray'))
  plt.show()
  # Output Part

• Block histogram

  1/4 사이즈를 갖는 블록 구간을 나누어 히스토그램을 출력하였다.

  tr, tc = r // 8, c // 8
  plt.figure(figsize=(20, 20))
  for i in range(3):
      for j in range(7):
          for k in range(7):
              plt.subplot2grid((21, 21), (14 + j, 7 * i + k))
              plt.hist(lbp_img[i, j * tr:(j + 2) * tr, k * tc:(k + 2) * tc],\
                       density=1, histtype='stepfilled', facecolor='k')
              plt.xticks([])
              plt.yticks([])

III. Result

• Input Image

  

• Output plot

  

• Histogram