Converting a PIL image to skimage?

Question

I have an image loaded in my code (very long and unnecessary, won't post here) that I need to work on with skimage to detect blobs in the image. However for some reason all of the images and attempts that I have used don't work.

The image is modified by PIL in the code above to make the objects I am trying to find only white (255, 255, 255), and all other pixels black. This code only needs to identify the positions of these. I have tried many different things to get this to work (converting them to a numpy array, not converting them to a numpy array ect.)

I am following these instructions : http://scikit-image.org/docs/dev/auto_examples/plot_blob.html for the basics, but it is not really a tutorial, more an example.

I think that the rgb2gray function is being problematic as it returns nothing but 0, whilst the print(numpy.array(img)) returns values between 0 and 1, suggesting that the loss of information is when the rgb2gray is called.

Here is the code that seems to be problematic:

img_gray = rgb2gray(numpy.array(img)) # Convert to numpy array for skimage
print(img_gray)
print(numpy.array(img))
img_blobs = blob_doh(img_gray, threshold=0.01, max_sigma=500)
print(img_blobs)

How can I fix this?

EDIT:

Here are what each of the images, arrays and lists print:

[[ 0.          0.          0.         ...,  0.          0.          0.        ] #img_gray
 [ 0.          0.00392157  0.01568627 ...,  0.          0.          0.        ]
 [ 0.          0.01176471  0.05882353 ...,  0.          0.          0.        ]
 ..., 
 [ 1.          0.99607843  0.96078431 ...,  0.          0.          0.        ]
 [ 1.          1.          0.98039216 ...,  0.          0.          0.        ]
 [ 1.          1.          0.99215686 ...,  0.          0.          0.        ]]
[[[  0   0   0] #img
  [  0   0   0]
  [  0   0   0]
  ..., 
  [  0   0   0] #img
  [  0   0   0]
  [  0   0   0]]

 [[  0   0   0] #img
  [  1   1   1]
  [  4   4   4]
  ..., 
  [  0   0   0] #img
  [  0   0   0]
  [  0   0   0]]

 [[  0   0   0] #img
  [  3   3   3]
  [ 15  15  15]
  ..., 
  [  0   0   0] #img
  [  0   0   0]
  [  0   0   0]]

 ..., 
 [[255 255 255] #img
  [254 254 254]
  [245 245 245]
  ..., 
  [  0   0   0] #img
  [  0   0   0]
  [  0   0   0]]

 [[255 255 255] #img
  [255 255 255]
  [250 250 250]
  ..., 
  [  0   0   0] #img
  [  0   0   0]
  [  0   0   0]]

 [[255 255 255]
  [255 255 255]
  [253 253 253]
  ..., 
  [  0   0   0]
  [  0   0   0]
  [  0   0   0]]]
[] #img_blobs

try creating the np.array(img) before rgb2gray() and print out what you've got before and after (as Matthew suggests check shape and dtype) Try making your own small test arrays (8,8,3) astype(np.uint8) etc to see if anything specific stops it working. — paddyg, Apr 28 '15 at 20:43
the max(img), min(img) and img.dtype all raise errors when running (Image type is not iterable and dtype is not an attribute of img) Creating the array before rgb2gray() does nothing, but in the edited part of the post it shows the testing printouts for (in order) img_gray, img, then img_blobs (the detected blobs). — Matthew Winfield, Apr 29 '15 at 15:06
After reading the printouts, img_gray appears to contain SOMETHING, but when using `PIL.Image.fromarray(img_gray).show()` I just get a black image. — Matthew Winfield, Apr 29 '15 at 15:13

score 1 · Answer 1 · answered Feb 12 '18 at 20:20

This will read image files (using PIL) into scimage:

from skimage import io
img = io.imread("./path/to/image.png")

Since you already had the images read and wanted to do the conversion of the data yourself you can look at how the plugin authors did it here: https://github.com/scikit-image/scikit-image/blob/master/skimage/io/_plugins/pil_plugin.py

Converting a PIL image to skimage?

1 Answers1