深度学习之自编码器实现——实现图像去噪

       大家好，我是带我去滑雪！

       自编码器是一种无监督学习的神经网络，是一种数据压缩算法，主要用于数据降维和特征提取。它的基本思想是将输入数据经过一个编码器映射到隐藏层，再通过一个解码器映射到输出层，使得输出层的结果与输入层的结果尽可能相似。自编码器的主要优点在于可以发现数据中的潜在模式和特征，进而用于特征提取或者压缩数据。它的主要应用领域包括图像去噪，识别和生成等。我们可以使用MLP或者CNN实现自编码器，在实战中自编码器可以应用在机器学习中的主成分分析，在保留主要特征下减少数据集的维数，也可以应用在CNN中，当处理尺寸很大的图片时，可以先使用自编码器来降维提取主要特征，使用主要特征进行学习。本期利用自编码器实现图像去噪。

目录

1、使用MLP创建自编码器

（1）导入相关模块和MNIST手写数据集

（2）数据预处理

（3）定义自编码器模型

（4）创建编码器模型

（5）创建解码器模型

（6）编译模型

（7）训练模型

 （8）使用自编码器来编码和解码手写数字图片

2、使用CNN创建自编码器

 （1）定义自编码器模型、编码器、解码器

（2）计算压缩图片和解码图片，画图展示

（3）使用CNN自编码器去除图片噪声

---

1、使用MLP创建自编码器

（1）导入相关模块和MNIST手写数据集

import numpy as np
import pandas as pd
from keras.datasets import mnist
import matplotlib.pyplot as plt
from keras.models import Model
from keras.layers import Input, Dense
 
# 载入数据集
(X_train, _), (X_test, _) = mnist.load_data()

（2）数据预处理

       首先将特征数据转化成28*28的向量，再将数据进行归一化：

#转换成 28*28 = 784 的向量
X_train = X_train.reshape(X_train.shape[0], 28*28).astype("float32")
X_test = X_test.reshape(X_test.shape[0], 28*28).astype("float32")
# 因为是固定范围, 所以执行归一化, 将 0~255 的灰度值转化成 0~1
X_train = X_train / 255
X_test = X_test / 255
 
X_train.shape

输出结果：

(60000, 784)

（3）定义编码器模型

input_img = Input(shape=(784,))
x = Dense(128, activation="relu")(input_img)
encoded = Dense(64, activation="relu")(x)
x = Dense(128, activation="relu")(encoded)
decoded = Dense(784, activation="sigmoid")(x)
autoencoder = Model(input_img, decoded)
autoencoder.summary()  # 显示模型信息

输出结果：

Model: "model"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_2 (InputLayer)        [(None, 784)]             0         
                                                                 
 dense_2 (Dense)             (None, 128)               100480    
                                                                 
 dense_3 (Dense)             (None, 64)                8256      
                                                                 
 dense_4 (Dense)             (None, 128)               8320      
                                                                 
 dense_5 (Dense)             (None, 784)               101136    
                                                                 
=================================================================
Total params: 218,192
Trainable params: 218,192
Non-trainable params: 0

（4）创建编码器模型

       编码器模型是自编码器模型的前半段。 

encoder = Model(input_img, encoded)
encoder.summary()    # 显示编码器模型信息

输出结果：

Model: "model_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_2 (InputLayer)        [(None, 784)]             0         
                                                                 
 dense_2 (Dense)             (None, 128)               100480    
                                                                 
 dense_3 (Dense)             (None, 64)                8256      
                                                                 
=================================================================
Total params: 108,736
Trainable params: 108,736
Non-trainable params: 0

（5）创建解码器模型

decoder_input = Input(shape=(64,))
decoder_layer = autoencoder.layers[-2](decoder_input)
decoder_layer = autoencoder.layers[-1](decoder_layer)
decoder = Model(decoder_input, decoder_layer)
decoder.summary()   # 显示解码器模型信息

输出结果：

Model: "model_2"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_3 (InputLayer)        [(None, 64)]              0         
                                                                 
 dense_4 (Dense)             (None, 128)               8320      
                                                                 
 dense_5 (Dense)             (None, 784)               101136    
                                                                 
=================================================================
Total params: 109,456
Trainable params: 109,456
Non-trainable params: 0
_________________________________________________________________

（6）编译模型

from tensorflow.keras  import optimizers
autoencoder.compile(loss="binary_crossentropy", optimizer= optimizers.Adam(),metrics=["accuracy"])

（7）训练模型

autoencoder.fit(X_train, X_train, validation_data=(X_test, X_test),epochs=30, batch_size=256, shuffle=True, verbose=1)

输出结果：

Epoch 23/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0748 - accuracy: 0.0131 - val_loss: 0.0743 - val_accuracy: 0.0131
Epoch 24/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0746 - accuracy: 0.0134 - val_loss: 0.0743 - val_accuracy: 0.0126
Epoch 25/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0745 - accuracy: 0.0130 - val_loss: 0.0739 - val_accuracy: 0.0142
Epoch 26/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0744 - accuracy: 0.0143 - val_loss: 0.0738 - val_accuracy: 0.0119
Epoch 27/30
235/235 [==============================] - 2s 6ms/step - loss: 0.0742 - accuracy: 0.0132 - val_loss: 0.0738 - val_accuracy: 0.0139
Epoch 28/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0741 - accuracy: 0.0139 - val_loss: 0.0738 - val_accuracy: 0.0149
Epoch 29/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0740 - accuracy: 0.0138 - val_loss: 0.0736 - val_accuracy: 0.0135
Epoch 30/30
235/235 [==============================] - 1s 6ms/step - loss: 0.0739 - accuracy: 0.0141 - val_loss: 0.0734 - val_accuracy: 0.0143

 （8）使用自编码器来编码和解码手写数字图片

# 压缩图片
encoded_imgs = encoder.predict(X_test)
 
# 还原图片
decoded_imgs = decoder.predict(encoded_imgs)
#  显示原始, 压缩和还原图片
n = 10  #绘制测试集的前10张图片
plt.figure(figsize=(20, 6))
for i in range(n):
    # 原始图片
    ax = plt.subplot(3, n, i + 1)
    ax.imshow(X_test[i].reshape(28, 28), cmap="gray")
    ax.axis("off")
    #压缩图片  
    ax = plt.subplot(3, n, i + 1 + n)
    ax.imshow(encoded_imgs[i].reshape(8, 8), cmap="gray")
    ax.axis("off")
    # 还原图片 
    ax = plt.subplot(3, n, i + 1 + 2*n)
    ax.imshow(decoded_imgs[i].reshape(28, 28), cmap="gray")
    ax.axis("off")
plt.savefig("E:\工作\硕士\博客\博客35-深度学习之自编码器实现——实现图像去噪/squares1.png",
            bbox_inches ="tight",
            pad_inches = 1,
            transparent = True,
            facecolor ="w",
            edgecolor ='w',
            dpi=300,
            orientation ='landscape')

输出结果：

2、使用CNN创建自编码器

 （1）定义自编码器模型、编码器、解码器

import numpy as np
from keras.datasets import mnist
from keras.models import Model
from keras.layers import Input, Conv2D, MaxPooling2D, UpSampling2D

(X_train, _), (X_test, _) = mnist.load_data()
#转换成4D 张量
X_train = X_train.reshape(X_train.shape[0], 28, 28, 1).astype("float32")
X_test = X_test.reshape(X_test.shape[0], 28, 28, 1).astype("float32")
X_train = X_train / 255
X_test = X_test / 255

# 定义 autoencoder 模型
input_img = Input(shape=(28,28,1))
x = Conv2D(16, (3,3), activation="relu", padding="same")(input_img)
x = MaxPooling2D((2,2), padding="same")(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(x)
x = MaxPooling2D((2,2), padding="same")(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(x)
encoded = MaxPooling2D((2,2), padding="same")(x)
 
x = Conv2D(8, (3,3), activation="relu", padding="same")(encoded)
x = UpSampling2D((2,2))(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(x)
x = UpSampling2D((2,2))(x)
x = Conv2D(16, (3,3), activation="relu")(x)
x = UpSampling2D((2,2))(x)
decoded = Conv2D(1, (3, 3), activation="sigmoid", padding="same")(x)
autoencoder = Model(input_img, decoded)
autoencoder.summary()     # 显示自编码器模型信息

#定义 编码器encoder 模型
encoder = Model(input_img, encoded)
encoder.summary()    #显示编码器模型信息

#定义解码器 decoder 模型
decoder_input = Input(shape=(4,4,8))
decoder_layer = autoencoder.layers[-7](decoder_input)
decoder_layer = autoencoder.layers[-6](decoder_layer)
decoder_layer = autoencoder.layers[-5](decoder_layer)
decoder_layer = autoencoder.layers[-4](decoder_layer)
decoder_layer = autoencoder.layers[-3](decoder_layer)
decoder_layer = autoencoder.layers[-2](decoder_layer)
decoder_layer = autoencoder.layers[-1](decoder_layer)
decoder = Model(decoder_input, decoder_layer)
decoder.summary()    # 显示解码器模型信息

输出结果：

Model: "model_6"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_6 (InputLayer)        [(None, 28, 28, 1)]       0         
                                                                 
 conv2d_14 (Conv2D)          (None, 28, 28, 16)        160       
                                                                 
 max_pooling2d_6 (MaxPooling  (None, 14, 14, 16)       0         
 2D)                                                             
                                                                 
 conv2d_15 (Conv2D)          (None, 14, 14, 8)         1160      
                                                                 
 max_pooling2d_7 (MaxPooling  (None, 7, 7, 8)          0         
 2D)                                                             
                                                                 
 conv2d_16 (Conv2D)          (None, 7, 7, 8)           584       
                                                                 
 max_pooling2d_8 (MaxPooling  (None, 4, 4, 8)          0         
 2D)                                                             
                                                                 
 conv2d_17 (Conv2D)          (None, 4, 4, 8)           584       
                                                                 
 up_sampling2d_6 (UpSampling  (None, 8, 8, 8)          0         
 2D)                                                             
                                                                 
 conv2d_18 (Conv2D)          (None, 8, 8, 8)           584       
                                                                 
 up_sampling2d_7 (UpSampling  (None, 16, 16, 8)        0         
 2D)                                                             
                                                                 
 conv2d_19 (Conv2D)          (None, 14, 14, 16)        1168      
                                                                 
 up_sampling2d_8 (UpSampling  (None, 28, 28, 16)       0         
 2D)                                                             
                                                                 
 conv2d_20 (Conv2D)          (None, 28, 28, 1)         145       
                                                                 
=================================================================
Total params: 4,385
Trainable params: 4,385
Non-trainable params: 0
_________________________________________________________________
Model: "model_7"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_6 (InputLayer)        [(None, 28, 28, 1)]       0         
                                                                 
 conv2d_14 (Conv2D)          (None, 28, 28, 16)        160       
                                                                 
 max_pooling2d_6 (MaxPooling  (None, 14, 14, 16)       0         
 2D)                                                             
                                                                 
 conv2d_15 (Conv2D)          (None, 14, 14, 8)         1160      
                                                                 
 max_pooling2d_7 (MaxPooling  (None, 7, 7, 8)          0         
 2D)                                                             
                                                                 
 conv2d_16 (Conv2D)          (None, 7, 7, 8)           584       
                                                                 
 max_pooling2d_8 (MaxPooling  (None, 4, 4, 8)          0         
 2D)                                                             
                                                                 
=================================================================
Total params: 1,904
Trainable params: 1,904
Non-trainable params: 0
_________________________________________________________________
Model: "model_8"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_7 (InputLayer)        [(None, 4, 4, 8)]         0         
                                                                 
 conv2d_17 (Conv2D)          (None, 4, 4, 8)           584       
                                                                 
 up_sampling2d_6 (UpSampling  (None, 8, 8, 8)          0         
 2D)                                                             
                                                                 
 conv2d_18 (Conv2D)          (None, 8, 8, 8)           584       
                                                                 
 up_sampling2d_7 (UpSampling  (None, 16, 16, 8)        0         
 2D)                                                             
                                                                 
 conv2d_19 (Conv2D)          (None, 14, 14, 16)        1168      
                                                                 
 up_sampling2d_8 (UpSampling  (None, 28, 28, 16)       0         
 2D)                                                             
                                                                 
 conv2d_20 (Conv2D)          (None, 28, 28, 1)         145       
                                                                 
=================================================================
Total params: 2,481
Trainable params: 2,481
Non-trainable params: 0
_________________________________________________________________

（2）计算压缩图片和解码图片，画图展示

# 压缩图片
encoded_imgs = encoder.predict(X_test)
#解压图片
decoded_imgs = decoder.predict(encoded_imgs)
 
# 显示原始, 压缩和还原图片
import matplotlib.pyplot as plt
 
n = 10  #显示测试集前10张图片
plt.figure(figsize=(20, 8))
for i in range(n):
    # 原始图片
    ax = plt.subplot(3, n, i + 1)
    ax.imshow(X_test[i].reshape(28, 28), cmap="gray")
    ax.axis("off")
    # 压缩图片
    ax = plt.subplot(3, n, i + 1 + n)
    ax.imshow(encoded_imgs[i].reshape(4, 4*8).T, cmap="gray")
    ax.axis("off")
    #  还原图片
    ax = plt.subplot(3, n, i + 1 + 2*n)
    ax.imshow(decoded_imgs[i].reshape(28, 28), cmap="gray")
    ax.axis("off")
plt.savefig("E:\工作\硕士\博客\博客35-深度学习之自编码器实现——实现图像去噪/squares2.png",
            bbox_inches ="tight",
            pad_inches = 1,
            transparent = True,
            facecolor ="w",
            edgecolor ='w',
            dpi=300,
            orientation ='landscape')

输出结果：

（3）使用CNN自编码器去除图片噪声

import numpy as np
from keras.datasets import mnist
from keras.models import Model
from keras.layers import Input, Conv2D, MaxPooling2D, UpSampling2D
 
(X_train, _), (X_test, _) = mnist.load_data()
#  转换成4D 张量
X_train = X_train.reshape(X_train.shape[0], 28, 28, 1).astype("float32")
X_test = X_test.reshape(X_test.shape[0], 28, 28, 1).astype("float32")
X_train = X_train / 255
X_test = X_test / 255
 
# 添加噪声 
nf = 0.5
size_train = X_train.shape
X_train_noisy = X_train+nf*np.random.normal(loc=0.0, scale=1.0,size=size_train)
X_train_noisy = np.clip(X_train_noisy, 0., 1.)
 
size_test = X_test.shape
X_test_noisy = X_test+nf*np.random.normal(loc=0.0,scale=1.0,size=size_test)
X_test_noisy = np.clip(X_test_noisy, 0., 1.)
 
# 定义自编码器 autoencoder 模型
input_img = Input(shape=(28,28,1))
x = Conv2D(16, (3,3), activation="relu", padding="same")(input_img)
x = MaxPooling2D((2,2), padding="same")(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(x)
x = MaxPooling2D((2,2), padding="same")(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(x)
encoded = MaxPooling2D((2,2), padding="same")(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(encoded)
x = UpSampling2D((2,2))(x)
x = Conv2D(8, (3,3), activation="relu", padding="same")(x)
x = UpSampling2D((2,2))(x)
x = Conv2D(16, (3,3), activation="relu")(x)
x = UpSampling2D((2,2))(x)
decoded = Conv2D(1, (3, 3), activation="sigmoid", padding="same")(x)
autoencoder = Model(input_img, decoded)
autoencoder.summary()   
 
# 定义编码器encoder 模型
encoder = Model(input_img, encoded)
encoder.summary()     
 
# 定义解码器 decoder 模型
decoder_input = Input(shape=(4,4,8))
decoder_layer = autoencoder.layers[-7](decoder_input)
decoder_layer = autoencoder.layers[-6](decoder_layer)
decoder_layer = autoencoder.layers[-5](decoder_layer)
decoder_layer = autoencoder.layers[-4](decoder_layer)
decoder_layer = autoencoder.layers[-3](decoder_layer)
decoder_layer = autoencoder.layers[-2](decoder_layer)
decoder_layer = autoencoder.layers[-1](decoder_layer)
decoder = Model(decoder_input, decoder_layer)
decoder.summary()   
 
#  编译模型
autoencoder.compile(loss="binary_crossentropy", optimizer="adam",metrics=["accuracy"])
# 训练模型
autoencoder.fit(X_train_noisy, X_train, 
                validation_data=(X_test_noisy, X_test), 
                epochs=10, batch_size=128, shuffle=True, verbose=2)
# 压缩图片
encoded_imgs = encoder.predict(X_test_noisy)
#解压缩图片
decoded_imgs = decoder.predict(encoded_imgs)
 
# 显示原始, 压缩和还原图片
import matplotlib.pyplot as plt
 
n = 10  
plt.figure(figsize=(20, 8))
for i in range(n):
    #  原始图片
    ax = plt.subplot(3, n, i + 1)
    ax.imshow(X_test_noisy[i].reshape(28, 28), cmap="gray")
    ax.axis("off")
    #压缩图片
    ax = plt.subplot(3, n, i + 1 + n)
    ax.imshow(encoded_imgs[i].reshape(4, 4*8).T, cmap="gray")
    ax.axis("off")
    # 还原图片
    ax = plt.subplot(3, n, i + 1 + 2*n)
    ax.imshow(decoded_imgs[i].reshape(28, 28), cmap="gray")
    ax.axis("off")
plt.savefig("E:\工作\硕士\博客\博客35-深度学习之自编码器实现——实现图像去噪/squares3.png",
            bbox_inches ="tight",
            pad_inches = 1,
            transparent = True,
            facecolor ="w",
            edgecolor ='w',
            dpi=300,
            orientation ='landscape')

输出结果：

---

更多优质内容持续发布中，请移步主页查看。

   点赞+关注,下次不迷路！