机器学习——图像训练

来源：cnblogs　　作者：抿嘴唇　　时间：2018/10/16 9:35:13　　对本文有异议

图片经过处理后图片会变成黑白无色彩的图像，但可以大概观察到图片中主体的轮廓信息，而还原后的图片的主体对象会被保留，图片中其他内容会变模糊，，主体对象得以突出，通过机器学习完成对图片的信息的提取，图片信息可以保存到本地像素查询本或数据库中

导入类库

1 import numpy as np
2 import cv2
3 import matplotlib.pyplot as plt
4 from sklearn.cluster import KMeans
5 from sklearn.utils import shuffle
6 from time import time
7 from skimage import io

提取和存储图像数据

 1 n_colors = 64
 2 # 读取图片像素数据
 3 tiger = cv2.imread("tiger.jpg")
 4 # print('tiger >>>>',tiger)
 5 login = cv2.imread('login.png')
 6 # 将图片像素数据标准化为0-1的数据并保存至数组中
 7 china = np.array(tiger, dtype=np.float64) / 255
 8 # print('china >>>>',china)
 9 w, h, d = original_shape = tuple(china.shape)
10 print('original_shape >>>>', original_shape)
11 print('w,h,d >>>>', w, h, d)  # w：层数，h行数，d列数
12 
13 image_array = np.reshape(china, (w * h, d))
14 # 每个点作为一个样本，维数为3，将三维数组china化为二维数组，列数不变，行数变为原行数乘层数
15 # print('image_array >>>>', image_array)
16 print('image_array shape>>>>', image_array.shape)

训练图像数据

1 t0 = time()
2 # 将所有点打乱顺序，取前1000个点
3 # 不使用所有点主要是为了训练模型的速度
4 image_array_sample = shuffle(image_array, random_state=0)[:1000]
5 
6 # 训练图片像素数据，将像素数据分为64类
7 # random_state = 0用来重现同样的结果，不设置则每次都是不同的随机结果
8 kmeans = KMeans(n_clusters=n_colors, random_state=0).fit(image_array_sample)
9 print("done in %0.3fs." % (time() - t0))  # 查看训练时间

预测

 1 print("Predicting color indices on the full image (k-means)")
 2 t0 = time()
 3 # 预测数据分类,image_array(921600,3)二维数组预测完毕后的结果labels是一维数组
 4 labels = kmeans.predict(image_array)
 5 print("done in %0.3fs." % (time() - t0))  # 查看预测时间
 6 print(labels)
 7 print(labels.shape)
 8 # 将labels从一维数组化为二维数组
 9 labels = labels.reshape(w, h)
10 print(labels)
11 print(labels.shape)

保存像素查询本和处理后的图像

1 def save_compress_data():
2     np.save('codebook_tiger.npy', kmeans.cluster_centers_)
3     io.imsave('compressed_tiger.png', labels)

还原图像

 1 def recreate_image(codebook, labels, w, h):
 2     # Recreate the (compressed) image from the code book & labels
 3     # 每个像素查询码本（对应0~63），取得其对应的像素值
 4     d = codebook.shape[1]
 5     image = np.zeros((w, h, d))
 6     label_idx = 0
 7     for i in range(w):
 8         for j in range(h):
 9             image[i][j] = codebook[labels[i, j]]
10             label_idx += 1
11     print('还原出的图像 >>>>', image)
12     return image

执行代码

 1 # save_compress_data()
 2 centers = np.load('codebook_tiger.npy')  # 像素查询码本
 3 c_image = io.imread('compressed_tiger.png')  # 这张图片里的像素已经过分类
 4 print('像素查询本 >>>>', centers)
 5 print(centers.shape)
 6 print(centers.shape[1])  # 0是行数，1是列数
 7 print('压缩的图像 >>>>', c_image)
 8 print(c_image.shape)
 9 
10 cv2.imshow("new", recreate_image(centers, c_image, w, h))
11 cv2.waitKey(0)

'''
取出压缩后图像的每一个数据即像素分类id(labels[i, j])，在像素查询本中查找该分类对应的三位像素一行数据(codebook[labels[i, j]])，
赋予新的image对象(无需指定列数，三位像素即3列)
original_shape >>>> (720, 1280, 3)
w,h,d >>>> 720 1280 3
image_array shape>>>> (921600, 3)
done in 0.583s.
Predicting color indices on the full image (k-means)
done in 0.740s.
[10 10 10 ... 60 60 60]
(921600,)
[[10 10 10 ... 42  8 42]
 [10 10 10 ... 42 42  8]
 [10 10 10 ... 42 42  8]
 ...
 [ 3  3 36 ... 60 60 60]
 [ 3  3 36 ... 60 60 60]
 [ 3  3 36 ... 60 60 60]]
(720, 1280)
像素查询本 >>>> [[0.26901961 0.27607843 0.35686275]
 [0.67189542 0.66887883 0.66747109]
 [0.09971989 0.09271709 0.10028011]
 [0.38901961 0.38184874 0.39383754]
 [0.83529412 0.8285205  0.83333333]
 [0.4402852  0.57468806 0.71764706]
 [0.46849673 0.49947712 0.47712418]
 [0.19622926 0.26651584 0.24494721]
 [0.30539216 0.45       0.60441176]
 [0.14444444 0.50359477 0.22581699]
 [0.64325609 0.63850267 0.62923351]
 [0.75148874 0.74524328 0.7405955 ]
 [0.18221289 0.18585434 0.20770308]
 [0.55294118 0.69233512 0.8631016 ]
 [0.09656863 0.28039216 0.43823529]
 [0.59155354 0.58924082 0.57797888]
 [0.53202614 0.38039216 0.7124183 ]
 [0.90756303 0.90364146 0.90140056]
 [0.3713555  0.46683717 0.54339301]
 [0.31215686 0.54352941 0.43372549]
 [0.23504902 0.32254902 0.30563725]
 [0.29694989 0.37342048 0.42461874]
 [0.19117647 0.32622549 0.4495098 ]
 [0.18431373 0.55757576 0.30516934]
 [0.70718954 0.76601307 0.83529412]
 [0.51328976 0.50544662 0.51568627]
 [0.36705882 0.23921569 0.61098039]
 [0.44416027 0.42983802 0.43631714]
 [0.10053476 0.16363636 0.23600713]
 [0.15022624 0.14434389 0.15806938]
 [0.40452489 0.50497738 0.59457014]
 [0.51265597 0.63030303 0.78324421]
 [0.29215686 0.30392157 0.30056022]
 [0.06876751 0.10294118 0.15644258]
 [0.0455243  0.05268542 0.06479113]
 [0.95294118 0.95294118 0.95803922]
 [0.34232026 0.35065359 0.33970588]
 [0.56254902 0.55431373 0.54196078]
 [0.2745098  0.49063181 0.27973856]
 [0.78676471 0.7814951  0.78112745]
 [0.21411765 0.20627451 0.43176471]
 [0.34196078 0.55843137 0.35803922]
 [0.36470588 0.5027451  0.66352941]
 [0.47189542 0.67973856 0.56078431]
 [0.49084967 0.55294118 0.63300654]
 [0.10889894 0.20301659 0.32488688]
 [0.65228758 0.78823529 0.92222222]
 [0.3372549  0.53411765 0.7427451 ]
 [0.71036415 0.70672269 0.69677871]
 [0.17019608 0.26431373 0.3696732 ]
 [0.39063181 0.44814815 0.40217865]
 [0.57303922 0.34656863 0.81617647]
 [0.28039216 0.37385621 0.49477124]
 [0.33440285 0.42816399 0.49411765]
 [0.43137255 0.61019608 0.81882353]
 [0.25743945 0.24705882 0.24313725]
 [0.27088989 0.34901961 0.36651584]
 [0.52990196 0.59215686 0.54166667]
 [0.39607843 0.55196078 0.51470588]
 [0.87511312 0.87179487 0.87722474]
 [0.41137255 0.46196078 0.46745098]
 [0.23627451 0.26470588 0.29362745]
 [0.1254902  0.44248366 0.18431373]
 [0.61265597 0.6197861  0.60463458]]
(64, 3)
3
压缩的图像 >>>> [[10 10 10 ... 42  8 42]
 [10 10 10 ... 42 42  8]
 [10 10 10 ... 42 42  8]
 ...
 [ 3  3 36 ... 60 60 60]
 [ 3  3 36 ... 60 60 60]
 [ 3  3 36 ... 60 60 60]]
(720, 1280)
还原出的图像 >>>> [[[0.64325609 0.63850267 0.62923351]
  [0.64325609 0.63850267 0.62923351]
  [0.64325609 0.63850267 0.62923351]
  ...
  [0.36470588 0.5027451  0.66352941]
  [0.30539216 0.45       0.60441176]
  [0.36470588 0.5027451  0.66352941]]
 
 [[0.64325609 0.63850267 0.62923351]
  [0.64325609 0.63850267 0.62923351]
  [0.64325609 0.63850267 0.62923351]
  ...
  [0.36470588 0.5027451  0.66352941]
  [0.36470588 0.5027451  0.66352941]
  [0.30539216 0.45       0.60441176]]
 
 [[0.64325609 0.63850267 0.62923351]
  [0.64325609 0.63850267 0.62923351]
  [0.64325609 0.63850267 0.62923351]
  ...
  [0.36470588 0.5027451  0.66352941]
  [0.36470588 0.5027451  0.66352941]
  [0.30539216 0.45       0.60441176]]
 
 ...
 
 [[0.38901961 0.38184874 0.39383754]
  [0.38901961 0.38184874 0.39383754]
  [0.34232026 0.35065359 0.33970588]
  ...
  [0.41137255 0.46196078 0.46745098]
  [0.41137255 0.46196078 0.46745098]
  [0.41137255 0.46196078 0.46745098]]
 
 [[0.38901961 0.38184874 0.39383754]
  [0.38901961 0.38184874 0.39383754]
  [0.34232026 0.35065359 0.33970588]
  ...
  [0.41137255 0.46196078 0.46745098]
  [0.41137255 0.46196078 0.46745098]
  [0.41137255 0.46196078 0.46745098]]
 
 [[0.38901961 0.38184874 0.39383754]
  [0.38901961 0.38184874 0.39383754]
  [0.34232026 0.35065359 0.33970588]
  ...
  [0.41137255 0.46196078 0.46745098]
  [0.41137255 0.46196078 0.46745098]
  [0.41137255 0.46196078 0.46745098]]]
 
'''