深度学习图像预处理之：对数据集中的每张图片的 R,G,B 各个通道进行减均值除标准差操作

from keras.datasets import cifar10

(x_train, y_train), (x_test, y_test) = cifar10.load_data()

# 如果不处理的话，每张图的每个通道都是 (0-255) 之间的取值
x_train[0]

array([[[ 59,  62,  63],
        [ 43,  46,  45],
        [ 50,  48,  43],
        ...,
        [158, 132, 108],
        [152, 125, 102],
        [148, 124, 103]],

       [[ 16,  20,  20],
        [  0,   0,   0],
        [ 18,   8,   0],
        ...,
        [123,  88,  55],
        [119,  83,  50],
        [122,  87,  57]],

       [[ 25,  24,  21],
        [ 16,   7,   0],
        [ 49,  27,   8],
        ...,
        [118,  84,  50],
        [120,  84,  50],
        [109,  73,  42]],

       ...,

       [[208, 170,  96],
        [201, 153,  34],
        [198, 161,  26],
        ...,
        [160, 133,  70],
        [ 56,  31,   7],
        [ 53,  34,  20]],

       [[180, 139,  96],
        [173, 123,  42],
        [186, 144,  30],
        ...,
        [184, 148,  94],
        [ 97,  62,  34],
        [ 83,  53,  34]],

       [[177, 144, 116],
        [168, 129,  94],
        [179, 142,  87],
        ...,
        [216, 184, 140],
        [151, 118,  84],
        [123,  92,  72]]], dtype=uint8)

x_train.shape

(50000, 32, 32, 3)

统计各个通道的均值

r_mean = x_train[:,:,:,0].mean()
r_mean

125.306918046875

g_mean = x_train[:,:,:,1].mean()
g_mean

122.950394140625

b_mean = x_train[:,:,:,2].mean()
b_mean

113.86538318359375

统计各个通道的标准差

r_std = x_train[:,:,:,0].std()
g_std = x_train[:,:,:,1].std()
b_std = x_train[:,:,:,2].std()
print(r_std,g_std,b_std)

62.993219278136884 62.08870764001421 66.70489964063091

减均值除标准差

mean = [125.307, 122.95, 113.865]
std = [62.9932, 62.0887, 66.7048]
for i in range(3):
    x_train[:,:,:,i] = (x_train[:,:,:,i] - mean[i]) / std[i]

总结

def normalize_preprocessing(x_train,x_test):
    x_train = x_train.astype('float32')
    x_test = x_test.astype('float32')
    mean = np.array([125.307, 122.95, 113.865])
    std  = np.array([62.9932, 62.0887, 66.7048])

    x_train = (x_train - mean) / std
    x_test = (x_test - mean) / std
    return x_train, x_test

(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train, x_test = color_preprocessing(x_train, x_test)

'''现在整个数据集的均值为 0 方差为 1 了'''
x_train[0]

array([[[-1.0526057e+00, -9.8166007e-01, -7.6253873e-01],
        [-1.3066014e+00, -1.2393559e+00, -1.0323844e+00],
        [-1.1954782e+00, -1.2071439e+00, -1.0623672e+00],
        ...,
        [ 5.1899254e-01,  1.4575927e-01, -8.7924674e-02],
        [ 4.2374417e-01,  3.3017330e-02, -1.7787321e-01],
        [ 3.6024526e-01,  1.6911339e-02, -1.6288179e-01]],

       [[-1.7352191e+00, -1.6581117e+00, -1.4071699e+00],
        [-1.9892148e+00, -1.9802315e+00, -1.7069985e+00],
        [-1.7034696e+00, -1.8513836e+00, -1.7069985e+00],
        ...,
        [-3.6622990e-02, -5.6290430e-01, -8.8247013e-01],
        [-1.0012191e-01, -6.4343429e-01, -9.5742726e-01],
        [-5.2497718e-02, -5.7901031e-01, -8.5248733e-01]],

       [[-1.5923465e+00, -1.5936877e+00, -1.3921785e+00],
        [-1.7352191e+00, -1.8674896e+00, -1.7069985e+00],
        [-1.2113529e+00, -1.5453697e+00, -1.5870671e+00],
        ...,
        [-1.1599664e-01, -6.2732828e-01, -9.5742726e-01],
        [-8.4247179e-02, -6.2732828e-01, -9.5742726e-01],
        [-2.5886920e-01, -8.0449420e-01, -1.0773586e+00]],

       ...,

       [[ 1.3127290e+00,  7.5778693e-01, -2.6782176e-01],
        [ 1.2016059e+00,  4.8398507e-01, -1.1972901e+00],
        [ 1.1539817e+00,  6.1283296e-01, -1.3172214e+00],
        ...,
        [ 5.5074203e-01,  1.6186525e-01, -6.5759879e-01],
        [-1.1002299e+00, -1.4809458e+00, -1.6020585e+00],
        [-1.1478541e+00, -1.4326278e+00, -1.4071699e+00]],

       [[ 8.6823660e-01,  2.5850120e-01, -2.6782176e-01],
        [ 7.5711352e-01,  8.0534868e-04, -1.0773586e+00],
        [ 9.6348500e-01,  3.3903116e-01, -1.2572558e+00],
        ...,
        [ 9.3173552e-01,  4.0345511e-01, -2.9780459e-01],
        [-4.4936597e-01, -9.8166007e-01, -1.1972901e+00],
        [-6.7161220e-01, -1.1266140e+00, -1.1972901e+00]],

       [[ 8.2061243e-01,  3.3903116e-01,  3.2006722e-02],
        [ 6.7773986e-01,  9.7441293e-02, -2.9780459e-01],
        [ 8.5236186e-01,  3.0681917e-01, -4.0274459e-01],
        ...,
        [ 1.4397268e+00,  9.8327076e-01,  3.9180091e-01],
        [ 4.0786946e-01, -7.9724602e-02, -4.4771886e-01],
        [-3.6622990e-02, -4.9848035e-01, -6.2761593e-01]]], dtype=float32)

import matplotlib.pyplot as plt

plt.imshow(x_train[0])

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).