Filtre de Canny terminé;

Ajout des fonctions utiles au filtre de Canny dans 'usefull_func.py'
Pas de modification important dans 'sobel.py'

.gitignore

@@ -2,3 +2,5 @@
 imageEngine/images/
 imageEngine/test/
 imageEngine/__pycache__/
+imageEngine/filters/__pycache__/
+imageEngine/test.py

imageEngine/filters/canny.py

@@ -1,39 +1,32 @@
-from usefull_func import *
+from copy import deepcopy
-from math import sqrt, atan2
+from filters.usefull_func import *
+from math import sqrt, atan2, pi
 
-#En_cours...
-def filtre_canny(img):
 
-    def norme_gradient(pixel1, pixel2):
+def filtreCanny(img, Th):
-        color_x = pixel1[0]
+    Tl = Th / 2
-        color_y = pixel2[0]
 
-        norm = round(sqrt(color_x**2 + color_y**2))
+    filtred_image = filtre_gaussien(img)
-        norm = min(norm, 255)
 
-        grad = atan2(color_y, color_x)
+    norme_gradient, angle_normale_gradient = calculGradient(filtred_image)
-        return norm, grad
 
-    def liste_normGrad(im1, im2):
+    non_maxima = dltNoMaxima(norme_gradient, angle_normale_gradient)
-        liste = []
-        for j in range(len(im1)):
-            ligne = []
-            for i in range(len(im1[0])):
-                normGrad = norme_gradient(im1[j][i], im2[j][i])
-                ligne.append(normGrad)
-            liste.append(ligne)
-        return liste
 
-    if not is_greyscale(img):
+    contours = seuillageHysteresis(non_maxima, angle_normale_gradient, Th, Tl)
-        img = greyscale(img)
 
-    mat_x = [[-1,0,1]]
+    return contours
-    mat_y = [[1],[0],[-1]]
 
-    #lissage/suppression des bri
+"""
-    img_no_bruit = convolution_gauss(img)
+def filtreCannySemiAuto(img, centile):
-    Jx = convolution(img, mat_x)
+    filtred_image = filtre_gaussien(img)
-    Jy = convolution(img, mat_y)
-    normGrad = liste_normGrad(Jx, Jy)
-    #Suppresion des non-maximum
 
+    norme_gradient, angle_normale_gradient = calculGradient(filtred_image)
+
+    non_maxima = dltNoMaxima(norme_gradient, angle_normale_gradient)
+
+    Th = calculTh(norme_gradient, centile)
+    Tl = Th / 2
+    contours = seuillageHysteresis(non_maxima, angle_normale_gradient, Th, Tl)
+
+    return contours
+"""

imageEngine/filters/sobel.py

@@ -1,5 +1,4 @@
-from usefull_func import *
+from filters.usefull_func import *
-
 
 def filtre_sobel(img):
 

imageEngine/filters/usefull_func.py

@@ -1,4 +1,5 @@
-from math import sqrt
+from copy import deepcopy
+from math import atan2, sqrt, pi
 
 def greyscale(mat_img):
     gray_img = []
@@ -10,26 +11,29 @@ def greyscale(mat_img):
         gray_img.append(lig)
     return gray_img
 
+
 def appliquer_convolution(img, mat, i, j):
     somme = 0
-    for y in range(len(mat)):
+    for x in range(len(mat)):
-        for x in range(len(mat[0])):
+        for y in range(len(mat[0])):
-            pixel_i = i - (len(mat[0]) // 2) + x
+            coord_i = i - (len(mat) // 2) + x
-            pixel_j = j - (len(mat) // 2) + y
+            corrd_j = j - (len(mat[0]) // 2) + y
-            pix = pixel(img, pixel_i, pixel_j)
+            pix = pixel(img, coord_i, corrd_j)
-            somme += pix[0]*mat[y][x]
+            somme += pix[0]*mat[x][y]
     return min(max(somme,0), 255)
 
+
 def convolution(mat_img, mat):
     return_img = []
-    for j in range(len(mat_img)):
+    for i in range(len(mat_img)):
         ligne = []
-        for i in range(len(mat_img[0])):
+        for j in range(len(mat_img[0])):
             val = appliquer_convolution(mat_img, mat, i, j)
             ligne.append((val,)*3)
         return_img.append(ligne)
     return return_img
 
+
 def is_greyscale(img):
     _greyscale = True
     for ligne in img:
@@ -41,6 +45,7 @@ def is_greyscale(img):
             break
     return _greyscale
 
+
 def invert(img):
     result_image = []
     for ligne in img:
@@ -50,25 +55,28 @@ def invert(img):
         result_image.append(result_ligne)
     return result_image
 
+
 def pixel(img, i, j, default=(0,0,0)):
     #i la colone et j la ligne
-    if 0 <= i < len(img[0]) and 0 <= j < len(img):
+    if 0 <= i < len(img) and 0 <= j < len(img[0]):
-        return img[j][i]
+        return img[i][j]
     else:
         return default
 
+
 def reduction_bruit(img, mat, i, j):
     somme = 0
-    for y in range(len(mat)):
+    for x in range(len(mat)):
-        for x in range(len(mat[0])):
+        for y in range(len(mat[0])):
-            pixel_i = i - (len(mat[0]) // 2) + x
+            pixel_i = i - (len(mat) // 2) + x
-            pixel_j = j - (len(mat) // 2) + y
+            pixel_j = j - (len(mat[0]) // 2) + y
             pix = pixel(img, pixel_i, pixel_j)
-            somme += pix[0]*mat[y][x]
+            somme += pix[0]*mat[x][y]
     normalise = round(somme)
     return normalise
 
-def convolution_gauss(mat_img):
+
+def filtre_gaussien(mat_img):
     mat_gauss = [
         [2/159, 4/159, 5/159, 4/159,2/159],
         [4/159, 9/159,12/159, 9/159,4/159],
@@ -78,28 +86,208 @@ def convolution_gauss(mat_img):
     ]
 
     return_img = []
-    for j in range(len(mat_img)):
+    for i in range(len(mat_img)):
         ligne = []
-        for i in range(len(mat_img[0])):
+        for j in range(len(mat_img[0])):
             val = reduction_bruit(mat_img, mat_gauss, i, j)
             ligne.append((val,)*3)
         return_img.append(ligne)
     return return_img
 
+
 def calcul_norme(pixel1, pixel2):
     valeur = pixel1[0]**2 + pixel2[0]**2
     norm = round(sqrt(valeur))
     norm = int(min(norm, 255))
     return norm
 
+
 def application_norme(im_x, im_y):
     result_image = []
-    for j in range(len(im_x)):
+    for i in range(len(im_x)):
         ligne = []
-        for i in range(len(im_x[0])):
+        for j in range(len(im_x[0])):
-            pixel1 = im_x[j][i]
+            pixel1 = im_x[i][j]
-            pixel2 = im_y[j][i]
+            pixel2 = im_y[i][j]
             norme = calcul_norme(pixel1, pixel2)
             ligne.append((norme,)*3)
         result_image.append(ligne)
     return result_image
+
+
+def calculGradient(filtred_image):
+    mask_x = [[1, 0, -1]]
+    mask_y = [[1],[0],[-1]]
+
+    mask_gradient_x = convolution(filtred_image, mask_x)
+    mask_gradient_y = convolution(filtred_image, mask_y)
+
+    norme_gradient = copyNullMatrix(filtred_image)
+    angle_normal_gradient = copyNullMatrix(filtred_image)
+
+    for i in range(len(filtred_image)):
+        for j in range(len(filtred_image[0])):
+            Jx = mask_gradient_x[i][j][0]
+            Jy = mask_gradient_y[i][j][0]
+            norme_gradient[i][j] = sqrt(Jx**2 + Jy**2)
+            angle_temp = atan2(Jy,Jx)
+            angle_normal_gradient[i][j] = transform_angle(angle_temp)
+    
+    return norme_gradient, angle_normal_gradient
+
+
+def copyNullMatrix(mat):
+    nullMat = deepcopy(mat)
+    for i in range(len(nullMat)):
+        for j in range(len(nullMat[0])):
+            nullMat[i][j] = 0
+    return nullMat
+
+
+def transform_angle(radient):
+    angle = radient * 180 / pi
+    if angle < 0:
+        angle += 180
+
+    #On veut que la valeur de l'angle soit 0, 45, 90 ou 135°
+    seuil_min_45 = 45/2
+    seuil_min_90 = (90+45)/2
+    seuil_min_135 = (135+90)/2
+    seuil_max_135 = (180+135)/2
+
+    if seuil_min_45 <= angle < seuil_min_90:
+        angle = 45
+    elif seuil_min_90 <= angle < seuil_min_135:
+        angle = 90
+    elif seuil_min_135 <= angle < seuil_max_135:
+        angle = 135
+    else:
+        angle = 0
+    
+    return angle
+
+
+def dltNoMaxima(norme_gradient, angle_normal_gradient):
+    non_maxima = copyNullMatrix(norme_gradient)
+
+    for i in range(len(non_maxima)):
+        for j in range(len(non_maxima[0])):
+            angle = angle_normal_gradient[i][j]
+            voisin1, voisin2 = norm_voisin(norme_gradient, angle, i, j)
+            if norme_gradient[i][j] < voisin1 or norme_gradient[i][j] < voisin2:
+                non_maxima[i][j] = 0
+            else:
+                non_maxima[i][j] = norme_gradient[i][j]
+    return non_maxima
+
+
+def get_norm(norm_list, i, j):
+    norm = 0
+    if 0 <= i < len(norm_list) and 0 <= j < len(norm_list[0]):
+        norm = norm_list[i][j]
+    return norm
+
+
+def norm_voisin(norm_list, angle, i, j):
+    voisin1 = None
+    voisin2 = None
+
+    if angle == 0:
+        voisin1 = get_norm(norm_list,i,j-1)
+        voisin2 = get_norm(norm_list,i,j+1)
+    elif angle == 45:
+        voisin2 = get_norm(norm_list,i-1,j+1)
+        voisin1 = get_norm(norm_list,i+1,j-1)
+    elif angle == 90:
+        voisin1 = get_norm(norm_list,i-1,j)
+        voisin2 = get_norm(norm_list,i+1,j)
+    elif angle == 135:
+        voisin2 = get_norm(norm_list,i-1,j-1)
+        voisin1 = get_norm(norm_list,i+1,j+1)
+    
+    return voisin1, voisin2
+
+
+def seuillageHysteresis(non_maxima, angle_normale_gradient, Th, Tl):
+    contours = deepcopy(non_maxima)
+
+    for i in range(len(angle_normale_gradient)):
+        for j in range(len(angle_normale_gradient[0])):
+            if non_maxima[i][j] > Th:
+                contours[i][j] = (255,)*3
+            elif non_maxima[i][j] < Tl:
+                contours[i][j] = (0,)*3
+    
+
+    for i in range(len(angle_normale_gradient)):
+        for j in range(len(angle_normale_gradient[0])):
+            if Tl <= non_maxima[i][j] <= Th:
+                angle = angle_normale_gradient[i][j] + 90
+                if angle >= 180:
+                    angle -= 180
+                
+                voisin1, voisin2 = norm_voisin(non_maxima, angle, i, j)
+                if voisin1 > Th and voisin2 > Th:
+                    contours[i][j] = (255,)*3
+                else:
+                    contours[i][j] = (0,)*3
+    return contours
+
+
+"""
+def calculTh(norme_gradient, centile):
+    histogramme, pas = calculHistogram(norme_gradient)
+    fonctionRepartition = calculFonctionRepartition(histogramme)
+    
+    nbrPixels = len(norme_gradient)*len(norme_gradient[0])
+    pivot = nbrPixels * centile
+
+    Th_index = 0
+    for i in range(len(fonctionRepartition)):
+        if (pivot - fonctionRepartition[0][Th_index]) > (pivot - fonctionRepartition[0][i]) and (pivot - fonctionRepartition[i] > 0):
+            Th_index = i
+    
+    Th = pas * (Th_index - 1)
+    return Th
+
+def calculHistogram(norme_gradient):
+    norme_max = Maxi(norme_gradient)
+    norme_min = Minim(norme_gradient)
+    ecart = norme_max - norme_min
+
+    nb_pas = 1000
+    pas = ecart / (nb_pas - 1)
+
+    histogram = [[0]*nb_pas]
+
+    for i in range(len(norme_gradient)):
+        for j in range(len(norme_gradient[0])):
+            valeur_pixel = norme_gradient[i][j]
+            position = floor((valeur_pixel - norme_min) / pas)
+            histogram[0][position] += 1
+    
+    return histogram, pas
+
+def calculFonctionRepartition(histogram):
+    fonctionRepartition = deepcopy(histogram)
+    for i in range(1, len(fonctionRepartition[0])):
+        fonctionRepartition[0][i] = fonctionRepartition[0][i-1] + histogram[0][i-1]
+
+    return fonctionRepartition
+
+def Maxi(mat):
+    mx = 0
+    for i in range(len(mat)):
+        for j in range(len(mat[0])):
+            if mat[i][j] > mx:
+                mx = mat[i][j]
+    return mx
+
+def Minim(mat):
+    mn = 1000
+    for i in range(len(mat)):
+        for j in range(len(mat[0])):
+            if mat[i][j] < mn:
+                mn = mat[i][j]
+    return mn
+"""