计算自定义图片数据集的平均像素值python代码

在计算机视觉领域对图片进行预处理时，通常需要先计算出数据集中所有图片的平均像素均值，本文详细给出了计算自定义图片数据集的平均像素的python代码。

1.获取自定义数据集的全部绝对路径

def is_imgfile(filepath):
    filepath = os.path.expanduser(filepath)
    if os.path.isfile(filepath) and imghdr.what(filepath):
        return True
    return False


def load_imgpaths_from_dir(dirpath, walk=False):
    imgpaths = []
    dirpath = os.path.expanduser(dirpath)
    if walk:  # wm,表示需要递归的读取文件夹
        for (root, _, files) in os.walk(dirpath):
            for file in files:
                file = os.path.join(root, file)
                if is_imgfile(file):  # wm,判断该文件是不是图片，如果是图片则把它的绝对路径加入list
                    imgpaths.append(file)
    else:
        for path in os.listdir(dirpath):
            path = os.path.join(dirpath, path)
            if not is_imgfile(path):
                continue
            imgpaths.append(path)
    return imgpaths

测试：

if __name__ == '__main__':
    path='G:\dataset\hymenoptera_data'
    img_path_list=load_imgpaths_from_dir(dirpath=path,walk=True)
    print(len(img_path_list))
    print(img_path_list)

结果：

2.计算图片数据集的三个通道的平均像素值

def compute_imgdataset_mean_pixel_value(imgset_all_path):
    '''
    imgset_all_path:是一个列表，包含着所有图片的绝对路径
    mpv：是三个通道的平均像素值
    '''
    mpv = np.zeros(shape=(3,))
    pbar = tqdm(total=len(imgset_all_path),desc='computing mean pixel value of training dataset...')
    for imgpath in imgset_all_path:
        img = Image.open(imgpath)
        x = np.array(img) / 255.
        mpv += x.mean(axis=(0, 1))#wm,对于图像的三个通道分别计算平均值，
        pbar.update()
    mpv /= len(imgset_all_path)
    pbar.close()
    return mpv

3.完整代码

import os
import numpy as np
import imghdr
from tqdm import tqdm
from PIL import Image


def is_imgfile(filepath):
    filepath = os.path.expanduser(filepath)
    if os.path.isfile(filepath) and imghdr.what(filepath):
        return True
    return False


def load_imgpaths_from_dir(dirpath, walk=False):
    imgpaths = []
    dirpath = os.path.expanduser(dirpath)
    if walk:  # wm,表示需要递归的读取文件夹
        for (root, _, files) in os.walk(dirpath):
            for file in files:
                file = os.path.join(root, file)
                if is_imgfile(file):  # wm,判断该文件是不是图片，如果是图片则把它的绝对路径加入list
                    imgpaths.append(file)
    else:
        for path in os.listdir(dirpath):
            path = os.path.join(dirpath, path)
            if not is_imgfile(path):
                continue
            imgpaths.append(path)
    return imgpaths

def compute_imgdataset_mean_pixel_value(imgset_all_path):
    '''
    imgset_all_path:是一个列表，包含着所有图片的绝对路径
    mpv：是三个通道的平均像素值
    '''
    mpv = np.zeros(shape=(3,))
    pbar = tqdm(total=len(imgset_all_path),desc='computing mean pixel value of training dataset...')
    for imgpath in imgset_all_path:
        img = Image.open(imgpath)
        x = np.array(img) / 255.
        mpv += x.mean(axis=(0, 1))#wm,对于图像的三个通道分别计算平均值，
        pbar.update()
    mpv /= len(imgset_all_path)
    pbar.close()
    return mpv

if __name__ == '__main__':
    path='G:\dataset\hymenoptera_data'
    img_path_list=load_imgpaths_from_dir(dirpath=path,walk=True)
    print(len(img_path_list))
    print(img_path_list)
    mvp=compute_imgdataset_mean_pixel_value(img_path_list)
    print("mvp:",mvp)

结果：