3

I'm reading images from filesystem using matplotlib imread function. However, it changes jpg image color when it displays those images. [Python 3.5, Anaconda3 4.3, matplotlib2.0]

# reading 5 color images of size 32x32
imgs_path = 'test_images'
test_imgs = np.empty((5,32,32,3), dtype=float)
img_names = os.listdir('test_images'+'/')
for i, img_name in enumerate(img_names):
    #reading in an image
    image = mpimg.imread(imgs_path+'/'+img_name)
    test_imgs[i] = image

#Visualize new raw images
plt.figure(figsize=(12, 7.5))
for i in range(5):
    plt.subplot(11, 4, i+1)
    plt.imshow(test_imgs[i]) 
    plt.title(i)
    plt.axis('off')
plt.show()

It is adding a bluish/greenish tint to all the images. Any mistake I'm doing?

smci
  • 32,567
  • 20
  • 113
  • 146
Shah-G
  • 652
  • 1
  • 8
  • 22

2 Answers2

2

matplotlib.image.imread or matplotlib.pyplot.imread read the image as unsigned integer array.
You then implicitely convert it to float.

matplotlib.pyplot.imshow interpretes arrays in both formats differently.

  • float arrays are interpreted between 0.0 (no color) and 1.0 (full color).
  • integer arrays are interpreted between 0 and 255.

The two options you have are thus:

  1. Use an integer array

    test_imgs = np.empty((5,32,32,3), dtype=np.uint8)

  2. divide the array by 255. prior to plotting:

    test_imgs = test_imgs/255.
ImportanceOfBeingErnest
  • 321,279
  • 53
  • 665
  • 712
  • I used: test_imgs[i] = (mpimg.imread(imgs_path+'/'+img_name)*255).astype('uint8') – Shah-G Feb 25 '17 at 19:45
  • I don't think that makes sense. imread gives you an image in integer type with values between 0 and 255. If you multiply that by 255 again, you'll end up with much too large values. – ImportanceOfBeingErnest Feb 25 '17 at 21:27
0

Matplotlib reads image in RGB format whereas if you use opencv it reads image in BGR format. First convert your .jpg image in RGB and then try displaying it. It worked for me.

Dhanashree Desai
  • 880
  • 6
  • 5