# How to suppress noises of this edge-detected image? [closed]

I've recently developed a new approach to detecting image edges. Everything seems to be normal except the noises! As you can see in the following image, there are many noises I'm not able to handle them properly. I used some bluring filters on the output, but the results were not as acceptable as those I expected. Moreover, since my digital number results generally range from 0.2e-8 to 0.5, thresholding methods, especially those of Python which only work on 8-bit images, have not been effective at all.

I would like to know your suggestions and comments on this denoising problem.

• Is this a gis related question, or is this related to just using the python image processing packages like openCV and PIL? – TsvGis Oct 11 '15 at 22:38

I think you can do some thresholding if you stretch your histogram. In the example below, I streched it between percentile 2 and 98 and set a treshold at 250. It looks like a start.

``````import numpy as np
import matplotlib.pyplot as plt
from scipy import ndimage as ndi
from skimage import exposure

# rebuilding your image from capure
pan = (noisy.sum(axis = 2)/4).astype('uint8') # fast flatening to 1 band

# stretching it bektween p2 and p98
p2, p98 = np.percentile(pan, (2, 98))
pan_stretch = exposure.rescale_intensity(pan, in_range=(p2, p98))

# arbitrary threshold
threshold = 250

# plotting

plt.subplot(2, 2, 1)
plt.imshow(pan, cmap='gray', interpolation='none')
plt.title('Original image')

plt.subplot(2,2,2)
plt.imshow(pan_stretch, cmap='gray', interpolation='none', vmin = threshold)
plt.title('Stetched image with Threshold at 250')

plt.subplot(2,2,3)
plt.hist(pan.ravel(), bins=256)

plt.subplot(2,2,4)
plt.hist(pan_stretch.ravel(), bins=256)
plt.axvline(threshold, color='r', linestyle='--', label='TRESHOLD')
plt.legend()
plt.show()
`````` If you also want to combine filters, you may have a look to scikit-image tools, maybe bilateral denoising `skimage.restoration.denoise_bilateral` which is an edge-preserving filter. See image denoising a picture