Merge pull request #2242 from grlee77/wavelet_nd

add n-dimensional support to denoise_wavelet

doc/examples/filters/plot_denoise.py

@@ -52,7 +52,8 @@
 ax[0, 2].imshow(denoise_bilateral(noisy, sigma_color=0.05, sigma_spatial=15))
 ax[0, 2].axis('off')
 ax[0, 2].set_title('Bilateral')
-ax[0, 3].imshow(denoise_wavelet(noisy, sigma=0.4*astro.std()))
+ax[0, 3].imshow(denoise_wavelet(noisy, sigma=0.4*astro.std(),
+                                multichannel=True))
 ax[0, 3].axis('off')
 ax[0, 3].set_title('Wavelet')
 
@@ -62,7 +63,8 @@
 ax[1, 2].imshow(denoise_bilateral(noisy, sigma_color=0.1, sigma_spatial=15))
 ax[1, 2].axis('off')
 ax[1, 2].set_title('(more) Bilateral')
-ax[1, 3].imshow(denoise_wavelet(noisy, sigma=0.6*astro.std()))
+ax[1, 3].imshow(denoise_wavelet(noisy, sigma=0.6*astro.std(),
+                                multichannel=True))
 ax[1, 3].axis('off')
 ax[1, 3].set_title('(more) Wavelet')
 ax[1, 0].imshow(astro)

skimage/restoration/_denoise.py

@@ -394,12 +394,13 @@ def _wavelet_threshold(img, wavelet, threshold=None, sigma=None, mode='soft'):
     return pywt.waverecn(denoised_coeffs, wavelet)
 
 
-def denoise_wavelet(img, sigma=None, wavelet='db1', mode='soft'):
+def denoise_wavelet(img, sigma=None, wavelet='db1', mode='soft',
+                    multichannel=False):
     """Performs wavelet denoising on an image.
 
     Parameters
     ----------
-    img : ndarray (2D/3D) of ints, uints or floats
+    img : ndarray ([M[, N[, ...P]][, C]) of ints, uints or floats
         Input data to be denoised. `img` can be of any numeric type,
         but it is cast into an ndarray of floats for the computation
         of the denoised image.
@@ -415,6 +416,9 @@ def denoise_wavelet(img, sigma=None, wavelet='db1', mode='soft'):
         An optional argument to choose the type of denoising performed. It
         noted that choosing soft thresholding given additive noise finds the
         best approximation of the original image.
+    multichannel : bool, optional
+        Apply wavelet denoising separately for each channel (where channels
+        correspond to the final axis of the array).
 
     Returns
     -------
@@ -457,16 +461,14 @@ def denoise_wavelet(img, sigma=None, wavelet='db1', mode='soft'):
 
     img = img_as_float(img)
 
-    if img.ndim not in {2, 3}:
-        raise ValueError('denoise_wavelet only supports 2D and 3D images')
-
-    if img.ndim == 2:
+    if multichannel:
+        out = np.empty_like(img)
+        for c in range(img.shape[-1]):
+            out[..., c] = _wavelet_threshold(img[..., c], wavelet=wavelet,
+                                             mode=mode, sigma=sigma)
+    else:
         out = _wavelet_threshold(img, wavelet=wavelet, mode=mode,
                                  sigma=sigma)
-    else:
-        out = np.dstack([_wavelet_threshold(img[..., c], wavelet=wavelet,
-                                            mode=mode, sigma=sigma)
-                         for c in range(img.ndim)])
 
     clip_range = (-1, 1) if img.min() < 0 else (0, 1)
     return np.clip(out, *clip_range)

skimage/restoration/tests/test_denoise.py

@@ -3,7 +3,7 @@
 
 from skimage import restoration, data, color, img_as_float, measure
 from skimage._shared._warnings import expected_warnings
-from skimage.measure import compare_ssim
+from skimage.measure import compare_psnr
 
 np.random.seed(1234)
 
@@ -310,16 +310,48 @@ def test_no_denoising_for_small_h():
 
 
 def test_wavelet_denoising():
-    for img in [astro_gray, astro]:
-        noisy = img.copy() + 0.1 * np.random.randn(*(img.shape))
+    for img, multichannel in [(astro_gray, False), (astro, True)]:
+        sigma = 0.1
+        noisy = img + sigma * np.random.randn(*(img.shape))
         noisy = np.clip(noisy, 0, 1)
-        # less energy in signal
-        denoised = restoration.denoise_wavelet(noisy, sigma=0.3)
-        assert denoised.sum()**2 <= img.sum()**2
 
-        # test changing noise_std (higher threshold, so less energy in signal)
-        assert (restoration.denoise_wavelet(noisy, sigma=0.2).sum()**2 <=
-                restoration.denoise_wavelet(noisy, sigma=0.1).sum()**2)
+        # Verify that SNR is improved when true sigma is used
+        denoised = restoration.denoise_wavelet(noisy, sigma=sigma,
+                                               multichannel=multichannel)
+        psnr_noisy = compare_psnr(img, noisy)
+        psnr_denoised = compare_psnr(img, denoised)
+        assert psnr_denoised > psnr_noisy
+
+        # Verify that SNR is improved with internally estimated sigma
+        denoised = restoration.denoise_wavelet(noisy,
+                                               multichannel=multichannel)
+        psnr_noisy = compare_psnr(img, noisy)
+        psnr_denoised = compare_psnr(img, denoised)
+        assert psnr_denoised > psnr_noisy
+
+        # Test changing noise_std (higher threshold, so less energy in signal)
+        res1 = restoration.denoise_wavelet(noisy, sigma=2*sigma,
+                                           multichannel=multichannel)
+        res2 = restoration.denoise_wavelet(noisy, sigma=sigma,
+                                           multichannel=multichannel)
+        assert (res1.sum()**2 <= res2.sum()**2)
+
+
+def test_wavelet_denoising_nd():
+    for ndim in range(1, 5):
+        # Generate a very simple test image
+        img = 0.2*np.ones((16, )*ndim)
+        img[[slice(5, 13), ] * ndim] = 0.8
+
+        sigma = 0.1
+        noisy = img + sigma * np.random.randn(*(img.shape))
+        noisy = np.clip(noisy, 0, 1)
+
+        # Verify that SNR is improved with internally estimated sigma
+        denoised = restoration.denoise_wavelet(noisy)
+        psnr_noisy = compare_psnr(img, noisy)
+        psnr_denoised = compare_psnr(img, denoised)
+        assert psnr_denoised > psnr_noisy
 
 
 if __name__ == "__main__":