Tensorflow专题7：图片&数据集预处理

图片处理

1 更改图片尺寸

import os
import os.path
from PIL import Image
'''
filein: 输入图片
fileout: 输出图片
width: 输出图片宽度
height:输出图片高度
type:输出图片类型（png, gif, jpeg...）
'''
def ResizeImage(filein, fileout, width, height, type):
    img = Image.open(filein)
    out = img.resize((width, height),Image.ANTIALIAS) #resize image with high-quality
    out.save(fileout, type)
if __name__ == "__main__":
    filein = r'image\test.png'
    fileout = r'image\testout.png'
    width = 60
    height = 85
    type = 'png'
    ResizeImage(filein, fileout, width, height, type)

2 把图片转为3维数组（两种方法）

用np + Image

import numpy as np
from PIL import Image
import matplotlib.pyplot as plt

image = Image.open(r'C:\Users\Administrator\Desktop\data\train\forest_001.jpg')   #读取图片文件
plt.imshow(image)
plt.show()            #将图片输出到屏幕

image_arr = np.array(image)      #将图片以数组的形式读入变量
print (image_arr)

用Tensorflow

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

image_contents = tf.read_file('D:/Tensorflow/slim/test images/000.jpg')   #读取文件

image = tf.image.decode_jpeg(image_contents, channels=3)     #解码jpeg

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    img=sess.run((image))          #img为三维数组
    print (img.shape)          #输出数组形状
    print (img)                     #打印数组

    plt.imshow(img)        #显示数组
    plt.show()

3 整合1，2实现把图片转为指定大小的三维数组★

import numpy as np
from PIL import Image
import matplotlib.pyplot as plt

def ImageToArray(imagePath, width, height):
    image = Image.open(imagePath, "r")   #读取图片文件
    plt.imshow(image)
    plt.show()            #将图片输出到屏幕
    out = image.resize((width, height),Image.ANTIALIAS)
    return np.array(out)   #将图片以数组的形式读入变量

image_arr = ImageToArray('D:/Tensorflow/slim/test images/000.jpg', 299, 299)
print (image_arr.shape)

数据处理

1 cifar-10 数据读取

def load_CIFAR_batch(filename):
    """ load single batch of cifar """
    with open(filename, 'rb') as f:
        datadict = pickle.load(f,encoding='latin1')
        X = datadict['data']
        Y = datadict['labels']
        X = X.reshape(10000, 3, 32,32).transpose(0,2,3,1).astype("float") # convert to channel_last
        Y = np.array(Y)
        return X, Y

#入口函数
def load_CIFAR10(ROOT):
    """ load all of cifar """
    xs = []
    ys = []
    for b in range(1,6):
        f = os.path.join(ROOT, 'data_batch_%d' % (b, ))
        X, Y = load_CIFAR_batch(f)
        xs.append(X)
        ys.append(Y)    
    X_train = np.concatenate(xs)#使变成行向量
    Y_train = np.concatenate(ys)
    del X, Y
    X_test, Y_test = load_CIFAR_batch(os.path.join(ROOT, 'test_batch'))
	f.close()
    return X_train[:5000], Y_train[:5000], X_test[:1000], Y_test[:1000]

2 标签数据 one-hot 编码

1、python方法

def GetOneHotLabel(label, numClass):
    m = np.zeros([len(label), numClass])
    for i in range(len(label)):
        m[i][label[i]] = 1
    return np.float64(m)

2、Tensorflow方法

#例子
one_hot_labels = tf.one_hot(indices=tf.cast(y, tf.int32), depth=_num_classes)

3 归一化

把图片像素压缩到[-1,1]之间

image = tf.cast(image, tf.float32) * (1. / 255) - 0.5

4 标准化

def per_image_standardization(image):
    image = image.astype('float32')
    image[:, :, 0] = (image[:, :, 0] - np.mean(image[:, :, 0])) / np.std(image[:, :, 0])
    image[:, :, 1] = (image[:, :, 1] - np.mean(image[:, :, 1])) / np.std(image[:, :, 1])
    image[:, :, 2] = (image[:, :, 2] - np.mean(image[:, :, 2])) / np.std(image[:, :, 2])
    return  image