CODEWITHSUNDEEP
pythonimagescipy
Shan
Shan

Reputation: 19243

The gray levels don't remain the same with scipy.misc.imresize()

I have an image which I would like to resize (upscale). I don't want extra gray levels to be introduced in my image. That's why I used the nearest neighbour interpolation as follows:

  scipy.misc.imresize(image, image2.shape, interp="nearest",mode="L")

Original gray levels in the image:

[ 0  2  4  5  8  9 10 11 12 14 15 16 17 18 19 20 21 22 23 25 26 27 28 29 30
31 32 35 36 37 38 41 43 45 46 47 51]

After interpolation:

[  0  10  20  25  40  45  50  55  60  70  75  80  85  90  95 100 105 110
115 125 130 135 140 145 150 155 160 175 180 185 190 205 215 225 230 235
255]

I also tried changing the mode, but it didn't help. I have no clue that how to fix it.

Upvotes: 2

Views: 780

Answers (1)

Warren Weckesser
Warren Weckesser

Reputation: 114781

imresize uses PIL or Pillow to do the actual work. It is the conversion to a PIL image with mode 'L' that triggers the rescaling of the data values. If the input data type is not 8 bit, the values are scaled to fill the 8 bit range.

One way to avoid this is to ensure that the input array has data type numpy.uint8. Then the values are not rescaled.

For example, here is a 3x4 image with 64 bit values (i.e. the data type of the array is numpy.int64):

In [132]: img
Out[132]: 
array([[ 1,  1,  2, 17],
       [ 4,  3,  1,  2],
       [ 1,  5,  4,  2]])

Here's what happens when this array is passed to imresize with mode='L':

In [133]: imresize(img, (6, 8), interp='nearest', mode='L')
Out[133]: 
array([[  0,   0,   0,   0,  16,  16, 255, 255],
       [  0,   0,   0,   0,  16,  16, 255, 255],
       [ 48,  48,  32,  32,   0,   0,  16,  16],
       [ 48,  48,  32,  32,   0,   0,  16,  16],
       [  0,   0,  64,  64,  48,  48,  16,  16],
       [  0,   0,  64,  64,  48,  48,  16,  16]], dtype=uint8)

If instead the input is first cast to np.uint8, the values are not rescaled:

In [134]: imresize(img.astype(np.uint8), (6, 8), interp='nearest', mode='L')
Out[134]: 
array([[ 1,  1,  1,  1,  2,  2, 17, 17],
       [ 1,  1,  1,  1,  2,  2, 17, 17],
       [ 4,  4,  3,  3,  1,  1,  2,  2],
       [ 4,  4,  3,  3,  1,  1,  2,  2],
       [ 1,  1,  5,  5,  4,  4,  2,  2],
       [ 1,  1,  5,  5,  4,  4,  2,  2]], dtype=uint8)

Upvotes: 5

Related Questions