【Keras】学习笔记13：Keras中的图像增强

一、增强前的图像

原始图像：

from keras.datasets import mnist
from matplotlib import pyplot as plt

# 从Keras导入Mnist数据集
(X_train, y_train), (X_validation, y_validation) = mnist.load_data()

# 显示9张手写数字的图片
for i in range(0, 9):
    plt.subplot(331 + i)
    plt.imshow(X_train[i], cmap=plt.get_cmap('gray'))

plt.show()

二、特征标准化

  对整个图像数据集的像素进行标准化，被称为特征标准化。能提高神经网络算法的性能。

X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')

  解释：训练数据集被构造为包括图像长宽的三维数组。对于多层感知器模型，我们必须将图像降维为像素矢量。在这种情况下，28×28大小的图像将变为784个像素的输入值。
我们可以使用NumPy数组上的reshape()函数轻松完成这个转换。我们还可以通过强制像素值的灰度值降低到为32（'float32）位来减少内存需求，原因之一是这是Keras默认的灰度值精度。

from keras.datasets import mnist
from keras.preprocessing.image import ImageDataGenerator
from matplotlib import pyplot as plt
from keras import backend

backend.set_image_data_format('channels_first')  #设置数据格式约定的值。

# 导入数据
(X_train, y_train), (X_validation, y_validation) = mnist.load_data()

X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')
X_validation = X_validation.reshape(X_validation.shape[0], 1, 28, 28).astype('float32')

# 图像特征化
imgGen = ImageDataGenerator(featurewise_center=True, featurewise_std_normalization=True)
imgGen.fit(X_train)

for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break

三、ZCA白化

  图像的白化处理是线性代数操作，能够减少图像像素矩阵的冗余和相关性。也就是更好的突出结构和特征给学习算法！两种比较流行的：主成分分析（PCA）和ZCA白化。ZCA白化转换后图像保持原始尺寸，被证明具有更好的适用性。

from keras.datasets import mnist
from keras.preprocessing.image import ImageDataGenerator
from matplotlib import pyplot as plt
from keras import backend

backend.set_image_data_format('channels_first')

# 导入数据
(X_train, y_train), (X_validation, y_validation) = mnist.load_data()

X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')
X_validation = X_validation.reshape(X_validation.shape[0], 1, 28, 28).astype('float32')

# ZCA白化
imgGen = ImageDataGenerator(zca_whitening=True)
imgGen.fit(X_train)
#flow 采集数据和标签数组，生成批量增强数据。
for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break

四、随机旋转、移动、剪切和反转图像

  对于样本不足，类似于样本增强？

from keras.datasets import mnist
from keras.preprocessing.image import ImageDataGenerator
from matplotlib import pyplot as plt
from keras import backend

backend.set_image_data_format('channels_first')

# 导入数据
(X_train, y_train), (X_validation, y_validation) = mnist.load_data()

X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')
X_validation = X_validation.reshape(X_validation.shape[0], 1, 28, 28).astype('float32')

# 图像旋转
imgGen = ImageDataGenerator(rotation_range=90)
imgGen.fit(X_train)

for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break

# 图像移动
imgGen = ImageDataGenerator(width_shift_range=0.2, height_shift_range=0.2)
imgGen.fit(X_train)

for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break

# 图像剪切
imgGen = ImageDataGenerator(shear_range=0.2)
imgGen.fit(X_train)

for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break

# 图像反转
imgGen = ImageDataGenerator(horizontal_flip=True, vertical_flip=True)
imgGen.fit(X_train)

for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break

五、保存增强后的图像

  使用keras可以实时的生成增强图像来训练模型，但是，当同时在多个模型的训练过程中使用这些增强图像时，若每次都通过实时生成，时间开销会非常大。因此，可以在训练的过程中将生成的图像保存到文件。

from keras.datasets import mnist
from keras.preprocessing.image import ImageDataGenerator
from matplotlib import pyplot as plt
from keras import backend
import os

backend.set_image_data_format('channels_first')

# 导入数据
(X_train, y_train), (X_validation, y_validation) = mnist.load_data()

X_train = X_train.reshape(X_train.shape[0], 1, 28, 28).astype('float32')
X_validation = X_validation.reshape(X_validation.shape[0], 1, 28, 28).astype('float32')

# ZCA白化
imgGen = ImageDataGenerator(zca_whitening=True)
imgGen.fit(X_train)

# 创建目录，并保存图像
try:
    os.mkdir('image')
except:
    print('The fold is exist!')
for X_batch, y_batch in imgGen.flow(X_train, y_train, batch_size=9, save_to_dir='image', save_prefix='oct',
                                    save_format='png'):
    for i in range(0, 9):
        plt.subplot(331 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break