Auto-Encoder的NOTMNIST数据集去噪

 1、import libraries 

import tensorflow as tf
import tensorflow.keras

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Conv2D, MaxPooling2D, UpSampling2D, Conv2DTranspose
from tensorflow.keras.constraints import max_norm


import matplotlib.pyplot as plt
import numpy as np
import gzip

2、下载并查看数据集

! wget -O train-images.gz https://github.com/davidflanagan/notMNIST-to-MNIST/raw/master/train-images-idx3-ubyte.gz
! wget -O test-images.gz https://github.com/davidflanagan/notMNIST-to-MNIST/raw/master/train-images-idx3-ubyte.gz
! wget -O train-labels.gz https://github.com/davidflanagan/notMNIST-to-MNIST/raw/master/train-labels-idx1-ubyte.gz
! wget -O test-labels.gz https://github.com/davidflanagan/notMNIST-to-MNIST/raw/master/t10k-labels-idx1-ubyte.gz

# function for extracting images
def image_data(filename, num_images):
    with gzip.open(filename) as f:
        f.read(16)
        buf = f.read(28 * 28 * num_images)
        data = np.frombuffer(buf, dtype=np.uint8).astype(np.float32)
        data = data.reshape(num_images, 28,28)
        return data

# function for extracting labels
def image_labels(filename, num_images):
    with gzip.open(filename) as f:
        f.read(8)
        buf = f.read(1 * num_images)
        labels = np.frombuffer(buf, dtype=np.uint8).astype(np.int64)
        return labels

# import train and test data
train_data = image_data('train-images.gz', 60000)
test_data = image_data('test-images.gz', 10000)

# import train and test labels
train_labels = image_labels('train-labels.gz', 60000)
test_labels = image_labels('test-labels.gz', 10000)

img_width, img_height = 28, 28
input_shape = (img_width, img_height, 1)
batch_size = 120
no_epochs = 50
validation_splits = 0.2
max_norm_value = 2.0
noise_factor = 0.5

label_dict = {i: a for i,a in zip(range(10), 'ABCDEFGHIJ')}

digits = [[train_data[idx], train_labels[idx]] for idx in
                        np.random.randint(len(train_data), size=50)]

plt.figure(figsize=(15,10))
for i in range(50):
  plt.subplot(5, 10, i+1%10)

  plt.imshow(digits[i][0])
  plt.title(label_dict[digits[i][1]], fontsize=18)
  plt.xticks([])
  plt.yticks([])

plt.show()

输出：

3、创建噪声图像

train_data = train_data.reshape(-1, 28,28, 1)
test_data = test_data.reshape(-1, 28,28, 1)

train_data = train_data / np.max(train_data)
test_data = test_data / np.max(test_data)

noise_train = train_data + noise_factor * np.random.normal(0,1,train_data.shape)
noise_test = test_data + noise_factor * np.random.normal(0, 1, test_data.shape)

fig, ax = plt.subplots(1,10)
fig.set_size_inches(20, 4)
for i in range(10):
    curr_img = np.reshape(train_data[i], (28,28))
    curr_lbl = train_labels[i]
    ax[i].imshow(curr_img)
    ax[i].set_xticks([])
    ax[i].set_yticks([])
plt.show()
fig, ax = plt.subplots(1,10)
fig.set_size_inches(20, 4)
for i in range(10):
    curr_img = np.reshape(noise_train[i], (28,28))
    curr_lbl = train_labels[i]
    ax[i].imshow(curr_img)
    ax[i].set_xticks([])
    ax[i].set_yticks([])
plt.show()

输出：

4、定义两个自动编码器

# Create the model
model = Sequential()
model.add(Conv2D(32, (3, 3),
                 activation='relu',
                 kernel_initializer='he_uniform',
                 padding='same',
                 input_shape=input_shape))
model.add(MaxPooling2D((2, 2),
                       padding='same'))
model.add(Conv2D(64, (3, 3),
                 activation='relu',
                 kernel_initializer='he_uniform',
                 padding='same'))
model.add(MaxPooling2D((2, 2),
                       padding='same'))
model.add(Conv2D(128, (3, 3),
                 activation='relu',
                 kernel_initializer='he_uniform',
                 padding='same'))


model.add(Conv2D(128, (3, 3),
                 activation='relu',
                 kernel_initializer='he_uniform',
                 padding='same'))
model.add(UpSampling2D((2, 2),
                       interpolation='bilinear'))

model.add(Conv2D(64, (3, 3),
                 activation='relu',
                 kernel_initializer='he_uniform',
                 padding='same'))
model.add(UpSampling2D((2, 2),
                       interpolation='bilinear'))
model.add(Conv2D(1, (3, 3),
                 activation='sigmoid',
                 padding='same'))

model.summary()


# model layers for autoencoder
model2 = Sequential()
model2.add(Conv2D(64, kernel_size=(3,3),
                 kernel_constraint=max_norm(max_norm_value),
                 activation = 'relu',
                 kernel_initializer= 'he_uniform',
                 input_shape = input_shape))
model2.add(Conv2D(32, kernel_size=(3,3),
                 kernel_constraint=max_norm(max_norm_value),
                 activation='relu',
                 kernel_initializer='he_uniform'))
model2.add(Conv2DTranspose(32,
                 kernel_size=(3,3),
                 kernel_constraint=max_norm(max_norm_value),
                 activation='relu',
                 kernel_initializer='he_uniform'))
model2.add(Conv2DTranspose(64, kernel_size=(3,3),
                 kernel_constraint= max_norm(max_norm_value),
                 activation='relu',
                 kernel_initializer= 'he_uniform'))
model2.add(Conv2D(1, kernel_size=(3,3),
                 kernel_constraint=max_norm(max_norm_value),
                 activation='sigmoid',
                 padding = 'same'))
model2.summary()

5、训练

model2.compile(optimizer='adam',
              loss = 'binary_crossentropy')
model2.fit(noise_train,
          train_data,
          validation_split= validation_splits,
          epochs=no_epochs,
          batch_size=batch_size)

注意model1用的是无参数的上采样，model2用的是转置卷积

6、去噪展示

fig_samples = noise_test[:10]
fig_original = test_data[:10]
fig_denoise = model2.predict(fig_samples)

for i in range(0, 10):
    noisy_img = noise_test[i]
    original_img = test_data[i]
    denoise_img = fig_denoise[i]

    fig, axes = plt.subplots(1, 3)
    fig.set_size_inches(6, 2.8)
    axes[0].imshow(noisy_img.reshape(28, 28))
    axes[0].set_xticks([])
    axes[0].set_yticks([])
    axes[0].set_title('Noisy')
    axes[1].imshow(original_img.reshape(28, 28))
    axes[1].set_xticks([])
    axes[1].set_yticks([])
    axes[1].set_title('Original')
    axes[2].imshow(denoise_img.reshape(28, 28))
    axes[2].set_xticks([])
    axes[2].set_yticks([])
    axes[2].set_title('Denoised')
plt.show()