图片数据集的均值与标准差计算_python中如何计算图像r、g、b、三个维度的均值和标准差

模型训练需要对图片数据集对样本进行归一化，因此要求均值和标准差
但因为笔记本内存不够，用数组寄存没办法实现，所以写了一个低配版的python脚本

1. 获取图片名称列表

def get_image_list(img_dir, isclasses=False):
    """将图像的绝对路径生成列表
    args: img_dir:存放图片的目录
          isclasses:图片是否按类别存放标志
    return: 图片文件名称列表
    """
    img_list = []
    # 路径下图像是否按类别分类存放
    if isclasses:
        img_file = os.listdir(img_dir)
        for class_name in img_file:
            if not os.path.isfile(os.path.join(img_dir, class_name)):     
                class_img_list = os.listdir(os.path.join(img_dir, class_name))
                img_list.extend(class_img_list)            
    else:
        img_list = os.listdir(img_dir)
    # print(img_list)
    print('image numbers: {}'.format(len(img_list)))
    return img_list
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

2. 求所有样本的R、G、B均值

def get_image_pixel_mean(img_dir, img_list, img_size):
    """求数据集图像的R、G、B均值
    args: img_dir:
          img_list:
          img_size:
    """
    R_sum = 0
    G_sum = 0
    B_sum = 0
    count = 0
    # 循环读取所有图片
    for img_name in img_list:
        img_path = os.path.join(img_dir, img_name)
        if not os.path.isdir(img_path):
            image = cv2.imread(img_path)
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            image = cv2.resize(image, (img_size, img_size))      # <class 'numpy.ndarray'>
            R_sum += image[:, :, 0].mean()
            G_sum += image[:, :, 1].mean()
            B_sum += image[:, :, 2].mean()
            count += 1
    R_mean = R_sum / count
    G_mean = G_sum / count
    B_mean = B_sum / count
    print('R_mean:{}, G_mean:{}, B_mean:{}'.format(R_mean,G_mean,B_mean))
    RGB_mean = [R_mean, G_mean, B_mean]
    return RGB_mean
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

3. 求所有样本的R、G、B标准差

def get_image_pixel_std(img_dir, img_mean, img_list, img_size):
    R_squared_mean = 0
    G_squared_mean = 0
    B_squared_mean = 0
    count = 0
    image_mean = np.array(img_mean)
    # 循环读取所有图片
    for img_name in img_list:
        img_path = os.path.join(img_dir, img_name)
        if not os.path.isdir(img_path):
            image = cv2.imread(img_path)
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            image = cv2.resize(image, (img_size, img_size))      # <class 'numpy.ndarray'>
            image = image - image_mean    # 零均值
            # 求单张图片的方差，并累加
            R_squared_mean += np.mean(np.square(image[:, :, 0]).flatten())
            G_squared_mean += np.mean(np.square(image[:, :, 1]).flatten())
            B_squared_mean += np.mean(np.square(image[:, :, 2]).flatten())
            count += 1
    # 求R、G、B的方差
    R_std = math.sqrt(R_squared_mean / count)
    G_std = math.sqrt(G_squared_mean / count)
    B_std = math.sqrt(B_squared_mean / count)
    print('R_std:{}, G_std:{}, B_std:{}'.format(R_std, G_std, B_std))
    RGB_std = [R_std, G_std, B_std]
    return RGB_std
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

3. 完整代码

import os
import cv2
import numpy as np
import math


def get_image_list(img_dir, isclasses=False):
    """将图像的名称列表
    args: img_dir:存放图片的目录
          isclasses:图片是否按类别存放标志
    return: 图片文件名称列表
    """
    img_list = []
    # 路径下图像是否按类别分类存放
    if isclasses:
        img_file = os.listdir(img_dir)
        for class_name in img_file:
            if not os.path.isfile(os.path.join(img_dir, class_name)):     
                class_img_list = os.listdir(os.path.join(img_dir, class_name))
                img_list.extend(class_img_list)         
    else:
        img_list = os.listdir(img_dir)
    print(img_list)
    print('image numbers: {}'.format(len(img_list)))
    return img_list


def get_image_pixel_mean(img_dir, img_list, img_size):
    """求数据集图像的R、G、B均值
    args: img_dir:
          img_list:
          img_size:
    """
    R_sum = 0
    G_sum = 0
    B_sum = 0
    count = 0
    # 循环读取所有图片
    for img_name in img_list:
        img_path = os.path.join(img_dir, img_name)
        if not os.path.isdir(img_path):
            image = cv2.imread(img_path)
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            image = cv2.resize(image, (img_size, img_size))      # <class 'numpy.ndarray'>
            R_sum += image[:, :, 0].mean()
            G_sum += image[:, :, 1].mean()
            B_sum += image[:, :, 2].mean()
            count += 1
    R_mean = R_sum / count
    G_mean = G_sum / count
    B_mean = B_sum / count
    print('R_mean:{}, G_mean:{}, B_mean:{}'.format(R_mean,G_mean,B_mean))
    RGB_mean = [R_mean, G_mean, B_mean]
    return RGB_mean


def get_image_pixel_std(img_dir, img_mean, img_list, img_size):
    R_squared_mean = 0
    G_squared_mean = 0
    B_squared_mean = 0
    count = 0
    image_mean = np.array(img_mean)
    # 循环读取所有图片
    for img_name in img_list:
        img_path = os.path.join(img_dir, img_name)
        if not os.path.isdir(img_path):
            image = cv2.imread(img_path)    # 读取图片
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            image = cv2.resize(image, (img_size, img_size))      # <class 'numpy.ndarray'>
            image = image - image_mean    # 零均值
            # 求单张图片的方差
            R_squared_mean += np.mean(np.square(image[:, :, 0]).flatten())
            G_squared_mean += np.mean(np.square(image[:, :, 1]).flatten())
            B_squared_mean += np.mean(np.square(image[:, :, 2]).flatten())
            count += 1
    R_std = math.sqrt(R_squared_mean / count)
    G_std = math.sqrt(G_squared_mean / count)
    B_std = math.sqrt(B_squared_mean / count)
    print('R_std:{}, G_std:{}, B_std:{}'.format(R_std, G_std, B_std))
    RGB_std = [R_std, G_std, B_std]
    return RGB_std


if __name__ == '__main__':
    image_dir = '/图片文件路径'
    image_list = get_image_list(image_dir, isclasses=False)
    RGB_mean = get_image_pixel_mean(image_dir, image_list, img_size=224)
    get_image_pixel_std(image_dir, RGB_mean, image_list, img_size=224)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88