使用from keras.preprocessing import image读取图片载入的是PIL,而cv2载入的图片通道为BGR,需要对其一一对应,进而使用模型对cv2读取的视频/图片进行预测
一、读取图片
import cv2
import PIL
import numpy as np
import keras.preprocessing
from matplotlib import pyplot as plt
path = 'data/court/train/tv/2.jpg'
img_height, img_width = 300, 400
# keras读取图片
img_keras = keras.preprocessing.image.load_img(path=path, target_size=(img_height, img_width))
plt.imshow(img_keras)
# opencv读取图片
img_cv2 = cv2.imread(path)
img_cv2 = PIL.Image.fromarray(cv2.cvtColor(img_cv2,cv2.COLOR_BGR2RGB))
img_cv2 = img_cv2.resize((img_width, img_height),PIL.Image.NEAREST)
plt.imshow(img_keras)
这里可以看到两张图片相差无几
二、图片转数组
# keras转数组
img_keras = keras.preprocessing.image.img_to_array(img_keras)
print(img_keras)
[[ 1. 0. 0.]
[ 0. 0. 7.]
[19. 22. 53.]
...
[23. 36. 78.]
[ 0. 9. 35.]
[ 0. 0. 0.]]
[[ 1. 3. 0.]
[ 0. 1. 0.]
[ 2. 5. 20.]
...
[ 1. 1. 25.]
[ 0. 0. 11.]
[ 0. 0. 0.]]
[[ 1. 3. 0.]
[ 0. 2. 0.]
[ 0. 0. 4.]
...
[ 9. 0. 7.]
[12. 2. 1.]
[ 0. 0. 0.]]]
# opencv转数组
img_cv2 = np.asarray(img_cv2, dtype='float32')
print(img_cv2)
[[ 1. 0. 0.]
[ 0. 0. 7.]
[19. 23. 52.]
...
[23. 37. 76.]
[ 0. 9. 35.]
[ 0. 0. 7.]]
[[ 1. 2. 0.]
[ 0. 1. 0.]
[ 2. 5. 22.]
...
[ 0. 2. 25.]
[ 0. 0. 11.]
[ 0. 0. 2.]]
[[ 1. 3. 0.]
[ 0. 2. 0.]
[ 0. 0. 5.]
...
[ 7. 0. 10.]
[11. 2. 5.]
[ 1. 0. 0.]]]
可以看到转数组后像素值相差无几,但是仍有区别,猜测是cv2转PIL的cvtColor方法造成的
三、数据预处理
# keras数据预处理
img_keras = np.expand_dims(img_keras, axis=0)
img_keras /= 255.
print(img_keras)
[[0.00392157 0.01176471 0. ]
[0. 0.00784314 0. ]
[0. 0. 0.01568628]
...
[0.03529412 0. 0.02745098]
[0.04705882 0.00784314 0.00392157]
[0. 0. 0. ]]]]
# opencv数据预处理
img_cv2 = np.expand_dims(img_cv2, axis=0)
img_cv2 /= 255.
print(img_cv2)
[[0.00392157 0.01176471 0. ]
[0. 0.00784314 0. ]
[0. 0. 0.01960784]
...
[0.02745098 0. 0.03921569]
[0.04313726 0.00784314 0.01960784]
[0.00392157 0. 0. ]]]]