CODEWITHSUNDEEP
numpyopencvscikit-image
dpetrini
dpetrini

Reputation: 1239

Convert signed float to uint8 image opencv

I have an image from cv2.matchTemplate that is float range -1,1: 

res = cv2.matchTemplate(img_gray,template,cv2.TM_CCOEFF_NORMED)

res has values like: [[ 0.00730964 -0.00275442 -0.02477949 ... -0.16014284 -0.13686109 -0.13015044]

I can see graycale map of pattern matching with:

cv2.imshow("Match", res)

However I want to see in colormap, using:

resC = cv2.applyColorMap(res, cv2.COLORMAP_JET)

Using this I immediately have issues like: "cv::ColorMap only supports source images of type CV_8UC1 or CV_8UC3 in function 'operator()'"

So I try skimage convertion:

from skimage import img_as_ubyte
res = img_as_ubyte(res)

or

from skimage import exposure
res = exposure.rescale_intensity(res, out_range=(0, 255))

With them I get outputs like: [[48 46 42 ... 14 19 20] [52 56 54 ... 22 28 30]

Better now, integers. However, something is wrong cause I get only (blue) monochrome colormaps, not the nice ones from cv2.COLORMAP_JET range. It seems is shifted somehow.

Any hints on how to convert from -1,1 to 0,255 properly?

Upvotes: 2

Views: 12932

Answers (1)

Onasafari
Onasafari

Reputation: 458

why this doesn't work:

I don't think this function is doing the rescaling you are hoping for. Consider the example from the reference manual below: 

>>> image = np.array([-10, 0, 10], dtype=np.int8)
>>> rescale_intensity(image, out_range=(0, 127))
array([  0,  63, 127], dtype=int8)

It maps the minimum number in the input array to 0 and the largest number to 1. If you don't have the exact values of -1, and 1 in your input array then using this function will not work.

what you can do instead:

I recommend writing a simple function to rescale the values from -1 to 1 into 0 to 255:

>>> image = np.random.uniform(-1,1,(3,3))
>>> scaled = (image + 1)*255/2.
>>> image
array([[ 0.59057256,  0.01683666, -0.24498247],
       [-0.25144806, -0.32312655, -0.02319944],
       [ 0.50878506, -0.04102033,  0.3094886 ]])
>>> scaled
array([[ 202.79800129,  129.64667417,   96.26473544],
       [  95.44037187,   86.3013643 ,  124.54207199],
       [ 192.37009459,  122.26990741,  166.95979601]])

How it works:

  • image + 1 shifts all numbers to the [0,2] range
  • (image +1)/2. scales all numbers to [0,1]
  • (image +1)*255/2. scales the numbers from [0,1] to [0,255]

Upvotes: 9

Related Questions