I'm facing an issue when opening a .tif using rasterio using the code below.
fp = 'image.tif'
image = rasterio.open(fp)
print(image.read())
When printing the content of the image, I get this
[[[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
...
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]]]
I verified all values and they are all 0. However, when dragging the image in QGIS, I can view it and confirm that the image contains values ranging from 101 to 122.
QGIS image
Any idea on how to read the image and get these 101 to 122 values as a numpy array ?
Here's a link to the image in question
Related
I have a numpy array of dimension 11*11 that I want to convert into an RGB image so I'm using this code :
import matplotlib.pyplot as plt
from PIL import Image as im
n_images = 1
train_data_np = train_data.to_numpy()
train_images = train_data_np[:n_images]
for i in range(n_images):
image_train = np.reshape(train_images[i], [11, 11])
image = im.fromarray(np.uint8(image_train))
plt.imshow(image)
plt.show()
My problem is that the image displayed is not all RGB because for this value :
[[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 255 0 0 150 25 43 7 43 0]
[ 0 0 12 0 0 255 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 0 0 0 255 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 0 0]
[ 0 0 0 0 0 0 0 0 0 255 0]]
it displayed this image
which doesn't respect RGB format as you can see where 0 it should be black instead
of purple and 255 should be white instead of yellow.
I tried to convert the numpy array of [11,11] into [11,11,3] to support RGB channels but it gave in the end a grayscale image only white and black. but it is not what I want. Here is the code that I used :
n_images = 1
train_data_np = train_data.to_numpy()
train_images = train_data_np[:n_images]
for i in range(n_images):
image_res = np.reshape(train_images[i],[11,11])
img2 = np.zeros( ( image_res.shape[0], image_res.shape[1], 3 ) )
img2[:,:,0] = image_res # same value in each channel
img2[:,:,1] = image_res
img2[:,:,2] = image_res
image_train = np.reshape(img2,[11,11,3])
image = im.fromarray(np.uint8(image_train),'RGB')
plt.imshow(image)
plt.show()
can someone explain to me how to implement or use a python function to transform the NumPy array 11x11 into an array of 11x11x3 using a colormap ?
This link contains an example of what i want really to do :
https://www.mathworks.com/help/matlab/ref/ind2rgb.html
Thank you in advance
I'm encoding my labels with label binarizer like this:
from sklearn.preprocessing import LabelBinarizer
# Transform labels to one-hot
lb = LabelBinarizer()
Y = lb.fit_transform(df.classification)
But when I print Y I get all zeros like:
[[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
...
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]]
I don't know if all the values in all rows are zeros or not. Unfortunately, I can't see the complete row and couldn't find a way to do so. Are these values right or not?
Any help would be appreciated.
I'm trying to produce a confusion matrix for 2 binary images. These are extracted (using binary thresholding) from 2 bands in a GeoTiff image, although I think this information should be irrelevant.
dataset = rasterio.open('NDBI.tif')
VH_26Jun2015 = dataset.read(1)
VH_30Sep2015 = dataset.read(3)
GND_Truth = dataset.read(7)
VH_diff = VH_26Jun2015 - VH_30Sep2015
ret,th1 = cv2.threshold(VH_diff,0.02,255,cv2.THRESH_BINARY)
print(confusion_matrix(GND_Truth,th1)
Error 1: I used the code above and ran into the problem mentioned here ValueError: multilabel-indicator is not supported for confusion matrix
I tried the argmax(axis=1) solution mentioned in the question and other places, but with a resulting 1983x1983 sized matrix. (This Error 1 is probably same as what the person in the question above ran into).
print(confusion_matrix(GND_Truth.argmax(axis=1),th1.argmax(axis=1)))
Output:
[[8 2 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
...
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]
[0 0 0 ... 0 0 0]]
I checked the contents of the GND_Truth and th1 and verified that they are binary.
numpy.unique(GND_Truth)
Output:
array([0., 1.], dtype=float32)
Error 2: Then I tried instead ravel() to flatten my binary images when passing to confusion_matrix like shown below, but resulting in a 3x3 matrix, whereas I'm expecting a 2x2 matrix.
print(confusion_matrix(GND_Truth.ravel().astype(int),th1.ravel().astype(int)))
Output:
[[16552434 0 2055509]
[ 6230317 0 1531602]
[ 0 0 0]]
Converting the data astype(int) did not really make a difference. Can you please suggest what might be causing these 2 errors?
I have a binary array of size 64x64x64, where a volume of 40x40x40 is set to "1" and rest is "0". I have been trying to rotate this cube about its center around z-axis using skimage.transform.rotate and also Opencv as:
def rotateImage(image, angle):
row, col = image.shape
center = tuple(np.array([row, col]) / 2)
rot_mat = cv2.getRotationMatrix2D(center, angle, 1.0)
new_image = cv2.warpAffine(image, rot_mat, (col, row))
return new_image
In the case of openCV, I tried, 2D rotation of each idividual slices in a cube (Cube[:,:,n=1,2,3...p]).
After rotating, total sum of the values in the array changes. This may be caused by interpolation during rotation. How can I rotate 3D array of this kind without adding anything to the array?
Ok so I understand now what you are asking. The closest I can come up with is scipy.ndimage. But there is a way interface with imagej from python if which might be easier. But here is what I did with scipy.ndimage:
from scipy.ndimage import interpolation
angle = 25 #angle should be in degrees
Rotatedim = interpolation.rotate(yourimage, angle, reshape = False,output = np.int32, order = 5,prefilter = False)
This worked for some angles to preserve the some and not others, perhaps by playing around more with the parameters you might be able to get your desired outcome.
One option is to convert into sparse, and transform the coordinates using a matrix rotation. Then transform back into dense. In 2 dimensions, this looks like:
import numpy as np
import scipy.sparse
import math
N = 10
space = np.zeros((N, N), dtype=np.int8)
space[3:7, 3:7].fill(1)
print(space)
print(np.sum(space))
space_coo = scipy.sparse.coo_matrix(space)
Coords = np.array(space_coo.nonzero()) - 3
theta = 30 * 3.1416 / 180
R = np.array([[math.cos(theta), math.sin(theta)], [-math.sin(theta), math.cos(theta)]])
space2_coords = R.dot(Coords)
space2_coords = np.round(space2_coords)
space2_coords += 3
space2_sparse = scipy.sparse.coo_matrix(([1] * space2_coords.shape[1], (space2_coords[0], space2_coords[1])), shape=(N, N))
space2 = space2_sparse.todense()
print(space2)
print(np.sum(space2))
Output:
[[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 1 1 1 1 0 0 0]
[0 0 0 1 1 1 1 0 0 0]
[0 0 0 1 1 1 1 0 0 0]
[0 0 0 1 1 1 1 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]]
16
[[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 1 0 0 0 0 0 0]
[0 0 1 1 1 1 0 0 0 0]
[0 0 1 1 1 1 1 0 0 0]
[0 1 1 0 1 1 0 0 0 0]
[0 0 0 1 1 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]]
16
The advantage is that you'll get exactly as many 1 values before and after the transform. The downsides is that you might get 'holes', as above, and/or duplicate coordinates, giving values of '2' in the final dense matrix.
I'm utilizing python GDAL to write a raster data into a .tif file. Here's the code:
import numpy, sys
from osgeo import gdal, utils
from osgeo.gdalconst import *
# register all of the GDAL drivers
gdal.AllRegister()
# open the image
inDs = gdal.Open("C:\\Documents and Settings\\patrick\\Desktop\\tiff elevation\\EBK1KM\\color_a1.tif",GDT_UInt16)
if inDs is None:
print "couldn't open input dataset"
sys.exit(1)
else:
print "opening was successful!"
cols = inDs.RasterXSize
rows = inDs.RasterYSize
bands = inDs.RasterCount
driver = inDs.GetDriver()
driver.Create("C:\\Documents and Settings\\patrick\\Desktop\\tiff elevation\\EBK1KM\\newfile.tif",cols,rows,3,GDT_UInt16)
outDs = gdal.Open("C:\\Documents and Settings\\patrick\\Desktop\\tiff elevation\\EBK1KM\\newfile.tif")
if outDs is None:
print "failure to create new file"
sys.exit(1)
outBand1 = outDs.GetRasterBand(1)
outBand2 = outDs.GetRasterBand(2)
outBand3 = outDs.GetRasterBand(3)
data1 = inDs.GetRasterBand(1).ReadAsArray()
data2 = inDs.GetRasterBand(2).ReadAsArray()
data3 = inDs.GetRasterBand(3).ReadAsArray()
outBand1.WriteArray(data1,0,0)
outBand2.WriteArray(data2,0,0)
outBand3.WriteArray(data3,0,0)
print "before closing out the file"
print outDs.GetRasterBand(1).ReadAsArray(700,700,5,5)
print outDs.GetRasterBand(2).ReadAsArray(700,700,5,5)
print outDs.GetRasterBand(3).ReadAsArray(700,700,5,5)
outDs.SetProjection(inDs.GetProjection())
outDs.SetGeoTransform(inDs.GetGeoTransform())
outDs = None
outDs = gdal.Open("C:\\Documents and Settings\\patrick\\Desktop\\tiff elevation\\EBK1KM\\newfile.tif")
print "after reopening"
print outDs.GetRasterBand(1).ReadAsArray(700,700,5,5)
print outDs.GetRasterBand(2).ReadAsArray(700,700,5,5)
print outDs.GetRasterBand(3).ReadAsArray(700,700,5,5)
The resultant output between the closing and reopening of the output dataset are different:
before closing out the file
[[ 36 35 55 121 0]
[ 54 0 111 117 0]
[ 0 117 152 56 0]
[ 89 122 56 0 0]
[102 107 0 25 53]]
[[ 68 66 126 200 0]
[ 78 0 166 157 0]
[ 0 235 203 70 0]
[229 251 107 0 0]
[241 203 0 42 121]]
[[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]]
after reopening
[[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]]
[[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]]
[[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]
[0 0 0 0 0]]
is there some command I'm missing to ensure that the file is written and saved prior to setting the variable to None? I've tried adding both of the following with no luck:
outband1.FlushCache()
outDs.FlushCache()
You don't need to Create then Open a raster (which you were reading GA_ReadOnly). You also don't need gdal.AllRegister() at the beginning, as it has already been called when you load GDAL into Python (see the Raster API tutorial).
Picking up somewhere above (with modifications):
# Create a new raster data source
outDs = driver.Create(out_fname, cols, rows, 3, gdal.GDT_UInt16)
# Write metadata
outDs.SetGeoTransform(inDs.GetGeoTransform())
outDs.SetProjection(inDs.GetProjection())
# Write raster data sets
for i in range(3):
outBand = outDs.GetRasterBand(i + 1)
outBand.WriteArray(data[i])
# Close raster file
outDs = None
Sometimes I add this to ensure the file is fully deallocated, and to prevent running into some gotchas:
del outDs, outBand