keras + tensorflow —— 图像处理

1. keras

• 预处理模块：from keras.preprocessing import image

  img_path = 'xx.jpg'
  img = image.load_img(img_path, target_size=(150, 150))
  	# 返回的是一个 PIL Image 实例
  	# target_size 仅用来指定 (img_width, img_height)
  img_array = image.img_to_array(img)
  img_array = img_array[None, :]
  				# (150, 150, 3) ⇒ (1, 150, 150, 3)
  img_array /= 255.
  plt.imshow(img_array[0])
  

• 图像增强生成器（generator）：

  from keras.preprocessing.image import ImageDataGenerator
  

2. tensorflow

• tf.image.peer_image_standardization

  raw_image = tf.reshape(raw_image, [image_depth, image_width, image_height])
  image = tf.transpose(raw_image, (1, 2, 0))
  image = tf.cast(image, tf.float32)
  
  # substract off the mean and divide by the variance
  image = tf.image.peer_image_standardization(image)