Merge branch 'master' into 'master'

Master

See merge request !22

src/methods/template/reconstruct.py → src/methods/quelletl/reconstruct.py

@@ -7,6 +7,12 @@ Students can call their functions (declared in others files of src/methods/your_
 import numpy as np
 
 from src.forward_model import CFA
+from src.methods.quelletl.some_function import interpolation
+import cv2
+from sklearn.ensemble import RandomForestRegressor
+from sklearn.model_selection import train_test_split
+from sklearn.metrics import mean_squared_error
+from scipy.signal import convolve2d
 
 
 def run_reconstruction(y: np.ndarray, cfa: str) -> np.ndarray:
@@ -19,13 +25,30 @@ def run_reconstruction(y: np.ndarray, cfa: str) -> np.ndarray:
     Returns:
         np.ndarray: Demosaicked image.
     """
-    # Performing the reconstruction.
-    # TODO
+
     input_shape = (y.shape[0], y.shape[1], 3)
     op = CFA(cfa, input_shape)
+    res = interpolation(y, op)
+    
+    return res
 
-    return np.zeros(op.input_shape)
+def quad_to_bayer(y):
+    for i in range(1, y.shape[0], 4):
+        save = np.copy(y[:,i])
+        y[:,i] = y[:,i+1]
+        y[:,i+1] = save
+    for j in range(1, y.shape[0], 4):
+        save = np.copy(y[j,:])
+        y[j,:] = y[j+1,:]
+        y[j+1,:] = save
+    for i in range(1, y.shape[0], 4):
+        for j in range(1, y.shape[0], 4):
+            save = np.copy(y[i,j])
+            y[i,j] = y[i+1,j+1]
+            y[i+1,j+1] = save
+    return y
 
+ker_bayer_green = np.array([[0, 1, 0], [1, 4, 1], [0, 1, 0]]) / 4
 
 ####
 ####

src/methods/quelletl/some_function.py

+import numpy as np
+
+def bayer_blue_red(res, y, i, j) : 
+    """
+    Compute estimated blue/red pixel in red/blue bayer pixel
+
+    Args : 
+        res     : estimated image
+        y       : image to reconstruct
+        i, j    : indices
+    Return : 
+        value   : value of the estimated pixels
+    """
+    K2 = res[i-1,j-1,1] - y[i-1,j-1]
+    K4 = res[i-1,j+1,1] - y[i-1,j+1]
+    K10 = res[i+1,j-1,1] - y[i+1,j-1]
+    K12 = res[i+1,j+1,1] - y[i+1,j+1]
+    value = res[i,j,1]-1/4*(K2+K4+K10+K12)
+    return value
+
+def bayer_green_vert(res, y, i, j) : 
+    """
+    Compute estimated blue/red pixel in green bayer pixel in vertical direction
+
+    Args : 
+        res     : estimated image
+        y       : image to reconstruct
+        i, j    : indices
+    Return : 
+        value   : value of the estimated pixels
+    """
+    k1 = res[i-1,j,1] - y[i-1,j]
+    k2 = res[i+1,j,1] - y[i+1,j]
+    value = y[i,j] - 1/2*(k1+k2)
+    return value
+
+def bayer_green_hor(res, y, i, j): 
+    """
+    Compute estimated blue/red pixel in green bayer pixel in horizontal direction
+
+    Args : 
+        res     : estimated image
+        y       : image to reconstruct
+        i, j    : indices
+    Return : 
+        value   : value of the estimated pixels
+    """
+    k1 = res[i,j-1,1] - y[i,j-1]
+    k2 = res[i,j+1,1] - y[i,j+1]
+    value = y[i,j] - 1/2*(k1+k2)
+    return value
+
+def interpolate_green(res, y, z):
+    """
+    Directional interpolation of the green channel
+
+    Args : 
+        res     : estimated image
+        y       : image to reconstruct
+        z       : bayer pattern
+    Return : 
+        res     : reconstructed image
+    """
+    for i in range(2,y.shape[0]-1):
+        for j in range(2,y.shape[1]-1):
+        # Vertical and horizontal gradient
+            if z[i,j,1]==0:
+                d_h = np.abs(y[i,j-1]-y[i,j+1])
+                d_v = np.abs(y[i-1,j]-y[i+1,j])
+                if d_h > d_v:
+                    green = (y[i-1,j]+y[i+1,j])/2
+                elif d_v > d_h:
+                    green = (y[i,j-1]+y[i,j+1])/2
+                else :
+                    green = (y[i,j-1]+y[i,j+1]+y[i-1,j]+y[i+1,j])/4
+            else : 
+                green = y[i,j]
+            res[i,j,1] = green
+    return res
+
+def quad_to_bayer(y):
+    """
+    Convert Quad Bayer to Bayer
+
+    Args : 
+        res     : estimated image
+        y       : image to reconstruct
+        i, j    : indices
+    Return : 
+        value   : value of the estimated pixels
+    """
+    for i in range(1, y.shape[0], 4):
+        save = np.copy(y[:,i])
+        y[:,i] = y[:,i+1]
+        y[:,i+1] = save
+    for j in range(1, y.shape[0], 4):
+        save = np.copy(y[j,:])
+        y[j,:] = y[j+1,:]
+        y[j+1,:] = save
+    for i in range(1, y.shape[0], 4):
+        for j in range(1, y.shape[0], 4):
+            save = np.copy(y[i,j])
+            y[i,j] = y[i+1,j+1]
+            y[i+1,j+1] = save
+    return y
+
+def interpolation(y, op):
+    """
+    Reconstruct image
+
+    Args : 
+        y       : image to reconstruct
+        op      : CFA operator
+    Return : 
+        np.ndarray: Demosaicked image.
+    """
+    if op.cfa == 'quad_bayer':
+        y = quad_to_bayer(y)
+        op.mask = quad_to_bayer(op.mask)
+
+    z = op.adjoint(y)
+    res = np.empty(op.input_shape)
+
+    # Interpolation of green channel
+    res = interpolate_green(res, y, z)
+    # Interpolation of R and B channels using channel correlation
+    for i in range(2,y.shape[0]-2):
+        for j in range(2, y.shape[1]-2):
+            # Bayer is Green
+            if z[i,j,1] != 0 : 
+                # Green is between 2 vertical bleu
+                if z[i+1,j,0] == 0:
+                    red = bayer_green_hor(res, y, i, j) # Compute Red channel
+                    blue = bayer_green_vert(res, y, i, j) # Compute Blue channel 
+                # Green is between 2 vertical red
+                else:
+                    blue = bayer_green_hor(res, y, i, j) # Compute Blue channel 
+                    red = bayer_green_vert(res, y, i, j) # Compute Red channel
+            # Bayer is red
+            elif z[i,j,0] != 0 :
+                red = y[i,j]    # Red channel
+                blue = bayer_blue_red(res, y, i, j) # Blue channel
+            # Bayer is bleue
+            elif z[i,j,2] != 0 :
+                blue = y[i,j] # Bleu channel
+                red = bayer_blue_red(res, y, i, j) # Res channel
+
+            res[i,j,0] = np.clip(red, 0, 255)
+            res[i,j,2] = np.clip(blue,0,255)
+    return res
+