def carv(I, nr, nc):
[n, m, d] = np.shape(I)
T = np.zeros(np.array([nr + 1, nc + 1]))
Ic = np.zeros(np.array([n - nr, m - nc, 3]))
table = np.empty((nr + 1, nc + 1), dtype=np.ndarray)
table[0, 0] = I
track = T
for i in range(0, nr + 1):
for j in range(0, nc + 1):
if (i == 0 and j == 0):
T[i, j] = 0
elif i == 0:
e = genEngMap(table[i, j - 1])
[Mx, Tbx] = cumMinEngVer(e)
[Ix, Ex] = rmVerSeam(table[i, j - 1], Mx, Tbx)
table[i, j] = Ix
T[i, j] = T[i, j - 1] + Ex
track[i, j] = -1
elif j == 0:
e = genEngMap(table[i - 1, j])
[My, Tby] = cumMinEngHor(e)
[Iy, Ey] = rmHorSeam(table[i - 1, j], My, Tby)
table[i, j] = Iy
T[i, j] = T[i - 1, j] + Ey
track[i, j] = 1
else:
e1 = genEngMap(table[i, j - 1])
e2 = genEngMap(table[i - 1, j])
[Mx, Tbx] = cumMinEngVer(e1)
[My, Tby] = cumMinEngHor(e2)
[Ix, Ex] = rmVerSeam(table[i, j - 1], Mx, Tbx)
[Iy, Ey] = rmHorSeam(table[i - 1, j], My, Tby)
if (T[i - 1, j] + Ey <= T[i, j - 1] + Ex):
table[i, j] = Iy
T[i, j] = T[i - 1, j] + Ey
track[i, j] = 1
else:
table[i, j] = Ix
T[i, j] = T[i, j - 1] + Ex
track[i, j] = -1
vid = []
[r, c] = [nr, nc]
while track[r, c] != 0:
if track[r, c] == 1:
vid.append(table[r, c])
r = r - 1
elif track[r, c] == -1:
vid.append(table[r, c])
c = c - 1
vid.append(table[0, 0])
vid.reverse()
Ic = table[nr, nc]
return Ic, T, vid
def carv(I, nr, nc, I_label):
# record=np.zeros(np.array([nr,nc]))
# T=mincost(nr, nc, I, record)
T = np.zeros(np.array([nr + 1, nc + 1]))
# x_mesh, y_mesh=np.meshgrid(np.arange(nr), np.arange(nc))
dict = {(0, 0): I}
#res_list = []
for i in range(0, nr + 1):
print(i)
for j in range(0, nc + 1):
print(j)
if (i == 0 and j == 0):
T[i, j] = 0
elif i == 0:
emap = genEngMap(dict[(i, j - 1)], I_label)
[Mx, Tbx] = cumMinEngVer(emap)
[Ix, Ex, I_label] = rmVerSeam(dict[(i, j - 1)], Mx, Tbx,
I_label)
dict[(i, j)] = Ix
T[i, j] = T[i, j - 1] + Ex
elif j == 0:
emap = genEngMap(dict[(i - 1, j)], I_label)
[My, Tby] = cumMinEngHor(emap)
[Iy, Ey, I_label] = rmHorSeam(dict[(i - 1, j)], My, Tby,
I_label)
dict[(i, j)] = Iy
T[i, j] = T[i - 1, j] + Ey
else:
emap = genEngMap(dict[(i, j - 1)], I_label)
emap2 = genEngMap(dict[(i - 1, j)], I_label)
[Mx, Tbx] = cumMinEngVer(emap)
[My, Tby] = cumMinEngHor(emap2)
[Ix, Ex, I_label] = rmVerSeam(dict[(i, j - 1)], Mx, Tbx,
I_label)
[Iy, Ey, I_label] = rmHorSeam(dict[(i - 1, j)], My, Tby,
I_label)
if (T[i - 1, j] + Ey <= T[i, j - 1] + Ex):
dict[(i, j)] = Iy
T[i, j] = T[i - 1, j] + Ey
else:
dict[(i, j)] = Ix
T[i, j] = T[i, j - 1] + Ex
#res_list.append(dict[(i, j)])
return dict[(i, j)], T #, res_list
def getTupleLeft(clone): dataEn = genEngMap(clone) Mx, Tbx = cumMinEngVer(dataEn) clone2, e = rmVerSeam(clone, Mx, Tbx) return clone2, e
def getTupleUp(clone): dataEn = genEngMap(clone) My, Tby = cumMinEngHor(dataEn) clone2, e = rmHorSeam(clone, My, Tby) return clone2, e
def do_the_actual_seam_carving(bounding_box):
e = genEngMap(bounding_box)
Mx, Tbx = cumMinEngVer(e)
seam_to_be_removed = rmVerSeam(bounding_box, Mx, Tbx)
return seam_to_be_removed
def carv(I, nr, nc):
res_list = []
height, width, color = I.shape
T = np.zeros((nr + 1, nc + 1))
e = genEngMap(I)
Ic = I
Mx, Tbx = cumMinEngVer(e)
My, Tby = cumMinEngHor(e)
removed_c = 0
removed_r = 0
T[0][1] = np.amin(Mx[height - 1])
My_transpose = np.transpose(My)
T[1][0] = np.amin(My_transpose[height - 1])
for i in range(nr + nc):
if removed_c == nc:
print("all columns removed")
My, Tby = cumMinEngHor(e)
Ic, E = rmHorSeam(Ic, My, Tby)
res_list.append(Ic)
e = genEngMap(Ic)
removed_r += 1
elif removed_r == nr:
print("all rows removed")
Mx, Tbx = cumMinEngVer(e)
Ic, E = rmVerSeam(Ic, My, Tby)
res_list.append(Ic)
e = genEngMap(Ic)
removed_c += 1
elif T[removed_r][removed_c +
1] > T[removed_r +
1][removed_c]: # here is the problem
print("removing row")
My, Tby = cumMinEngHor(e)
Ic, E = rmHorSeam(Ic, My, Tby)
res_list.append(Ic)
e = genEngMap(Ic)
removed_r += 1
Mx, Tbx = cumMinEngVer(e)
My, Tby = cumMinEngHor(e)
T[removed_r][removed_c + 1] = np.amin(Mx[height - 1 - removed_r])
My_transpose = np.transpose(My)
if (removed_r < nr):
T[removed_r + 1][removed_c] = np.amin(My_transpose[width - 1 -
removed_c])
else:
print("removing column")
Mx, Tbx = cumMinEngVer(e)
Ic, E = rmVerSeam(Ic, Mx, Tbx)
res_list.append(Ic)
e = genEngMap(Ic)
removed_c += 1
Mx, Tbx = cumMinEngVer(e)
My, Tby = cumMinEngHor(e)
if (removed_c < nc):
T[removed_r][removed_c + 1] = np.amin(Mx[height - 1 -
removed_r])
My_transpose = np.transpose(My)
T[removed_r + 1][removed_c] = np.amin(My_transpose[width - 1 -
removed_c])
return np.uint8(Ic), T
def carv(I, nr, nc):
res_list = []
print I.shape
height, width, color = I.shape
T = np.zeros((nr + 1, nc + 1))
e = genEngMap(I)
Ic = I
Mx, Tbx = cumMinEngVer(e)
My, Tby = cumMinEngHor(e)
removed_c = 0
removed_r = 0
T[0][1] = np.amin(Mx[height - 1])
My_transpose = np.transpose(My)
T[1][0] = np.amin(My_transpose[height - 1])
for i in range(nr + nc):
# print i
if removed_c == nc:
print "all columns removed"
My, Tby = cumMinEngHor(e)
Ic, E = rmHorSeam(Ic, My, Tby)
res_list.append(Ic)
e = genEngMap(Ic)
removed_r += 1
elif removed_r == nr:
print "all rows removed"
Mx, Tbx = cumMinEngVer(e)
Ic, E = rmVerSeam(Ic, My, Tby)
res_list.append(Ic)
e = genEngMap(Ic)
removed_c += 1
elif T[removed_r][removed_c +
1] > T[removed_r +
1][removed_c]: #here is the problem
print "romoving row"
My, Tby = cumMinEngHor(e)
Ic, E = rmHorSeam(Ic, My, Tby)
res_list.append(Ic)
e = genEngMap(Ic)
removed_r += 1
Mx, Tbx = cumMinEngVer(e)
My, Tby = cumMinEngHor(e)
T[removed_r][removed_c + 1] = np.amin(Mx[height - 1 - removed_r])
My_transpose = np.transpose(My)
if (removed_r < nr):
T[removed_r + 1][removed_c] = np.amin(My_transpose[width - 1 -
removed_c])
else:
print "removing column"
Mx, Tbx = cumMinEngVer(e)
Ic, E = rmVerSeam(Ic, Mx, Tbx)
res_list.append(Ic)
e = genEngMap(Ic)
removed_c += 1
Mx, Tbx = cumMinEngVer(e)
My, Tby = cumMinEngHor(e)
if (removed_c < nc):
T[removed_r][removed_c + 1] = np.amin(Mx[height - 1 -
removed_r])
My_transpose = np.transpose(My)
T[removed_r + 1][removed_c] = np.amin(My_transpose[width - 1 -
removed_c])
# print Ic.shape
imageio.mimsave('./Shanghai2Output.gif', res_list)
return Ic, T
# test = np.array([[5.5, 8, 4.5, 6, 3], [13, 9, 30, 27, 19], [6.5, 7, 6, 12.5, 8], [16, 24, 27, 11, 13], [3, 6, 4.5, 8, 5.5]])
# data1 = np.array([[[1, 2, 3], [1, 2, 3], [1, 6, 3], [1, 2, 3], [1, 2, 3]], [[4, 5, 6], [5, 5, 6], [4, 5, 6], [4, 5, 6], [4, 5, 6]], [[1, 2, 3], [1, 2, 3], [1, 2, 4], [1, 2, 3], [1, 2, 3]], [[4, 5, 6], [4, 5, 6], [4, 5, 6], [4, 7, 6], [4, 5, 6]], [[7, 8, 9], [7, 8, 9], [10, 11, 12], [3, 5, 0], [2, 1, 6]]])
# carv(data1, 2, 2)
# img1 = Image.open("shanghai2.jpg")
# data1 = np.asarray(img1, dtype="int32")
# Ic, T = carv(data1, 50, 30)
# Ic = np.asarray(Ic, dtype="uint8")
# plt.imshow(Ic)
# plt.show()
def carv(I, nr, nc):
# Your Code Here
N, M, _ = I.shape
# construct the empty matrix T
T = np.zeros((nr + 1, nc + 1))
# initialize the mapping dictionary
image_map = {}
image_map[(0, 0)] = I
# initial the track table, 1 is left and 2 is up
Tbt = np.zeros((nr + 1, nc + 1))
Tbt[0, :] = 1
Tbt[:, 0] = 2
Tbt[0, 0] = 0
# fill the first row and col of the T matrix
for i in range(1, nr + 1):
# new coor
coor = (i, 0)
up_coor = (i - 1, 0)
I_up = image_map[up_coor]
# new image with one row deleted
e = genEngMap(I_up)
My, Tby = cumMinEngHor(e)
Iy, E = rmHorSeam(I_up, My, Tby)
# add the new image to the dict and fill the (i,0) of E matrix
image_map[coor] = Iy
T[coor] = T[up_coor] + E
for j in range(1, nc + 1):
# new coor
coor = (0, j)
left_coor = (0, j - 1)
I_left = image_map[left_coor]
# new image with one column deleted
e = genEngMap(I_left)
Mx, Tbx = cumMinEngVer(e)
Ix, E = rmVerSeam(I_left, Mx, Tbx)
# add the new image to the dict and fill the (0,j) of E matrix
image_map[coor] = Ix
T[coor] = T[left_coor] + E
# DP to complete T matrix
for ii in range(1, nr + 1):
for jj in range(1, nc + 1):
coor = (ii, jj)
up_coor = (ii - 1, jj)
left_coor = (ii, jj - 1)
candidate_img_list = []
cost_list = []
# grow from left, delete col
I_left = image_map[left_coor]
e = genEngMap(I_left)
Mx, Tbx = cumMinEngVer(e)
Ix, Ex = rmVerSeam(I_left, Mx, Tbx)
# Ix = np.pad(Ix, ((0, ii), (0, jj)), 'constant')
cost_from_left = Ex + T[left_coor]
# grow from up, delete row
I_up = image_map[up_coor]
e = genEngMap(I_up)
My, Tby = cumMinEngHor(e)
Iy, Ey = rmHorSeam(I_up, My, Tby)
# Iy = np.pad(Iy, ((0,ii),(0,jj)), 'constant')
cost_from_up = Ey + T[up_coor]
# construct cost list (2 elements)
cost_list.append(cost_from_left)
cost_list.append(cost_from_up)
# add candidate imgs to the list
candidate_img_list.append(Ix)
candidate_img_list.append(Iy)
# construct Tbt and T table
# 1 is left and 2 is up
indentify_direction = np.argmin(cost_list)
Tbt[ii, jj] = indentify_direction + 1
T[ii, jj] = cost_list[indentify_direction]
image_map[coor] = candidate_img_list[indentify_direction]
# from (nr,nc) trace back along the shortest path
path, path_imags = shortest_path_delete_imags(image_map, Tbt)
# for the return Ic image
Ic = path_imags[-1]
# make the gif
make_gif(N, M, path_imags)
return Ic, T
def delVer(I):
e = genEngMap(I)
Mv, Tbv = cumMinEngVer(e)
Iv, Ev = rmVerSeam(I, Mv, Tbv)
return Iv, Ev
def delHor(I):
e = genEngMap(I)
Mh, Tbh = cumMinEngHor(e)
Ih, Eh = rmHorSeam(I, Mh, Tbh)
return Ih, Eh
def cutV(I):
Mx, Tbx = cumMinEngVer(genEngMap(I))
Ix, Ex = rmVerSeam(I, Mx, Tbx)
return Ix, Ex
def cutH(I):
My, Tby = cumMinEngHor(genEngMap(I))
Iy, Ey = rmHorSeam(I, My, Tby)
return Iy, Ey
def carv(I, nr, nc):
# Your Code Here
r = nr + 1
c = nc + 1
image_list = [[[] for j in range(c)] for i in range(r)]
T = np.zeros((r, c))
Tr = np.zeros((r, c))
image_list[0][0] = np.copy(I)
for i in list(range(1, c)):
img = np.copy(image_list[0][i - 1])
e = genEngMap.genEngMap(img)
M_left, Tb_left = cumMinEngVer.cumMinEngVer(e)
image, cost = rmVerSeam.rmVerSeam(img, M_left, Tb_left)
image_list[0][i] = np.copy(image)
Tr[0][i] = Tr[0][i - 1] + cost
for j in list(range(1, r)):
img = np.copy(image_list[j - 1][0])
e = genEngMap.genEngMap(img)
M_up, Tb_up = cumMinEngHor.cumMinEngHor(e)
image_b, cost = rmHorSeam.rmHorSeam(img, M_up, Tb_up)
image_list[j][0] = np.copy(image_b)
Tr[j][0] = Tr[j - 1][0] + cost
T[j][0] = 1
for i in list(range(1, r)):
for j in list(range(1, c)):
img = np.copy(image_list[i][j - 1])
e_left = genEngMap.genEngMap(img)
M_left, Tb_left = cumMinEngVer.cumMinEngVer(e_left)
image_left, cost_left = rmVerSeam.rmVerSeam(img, M_left, Tb_left)
img = np.copy(image_list[i - 1][j])
e_up = genEngMap.genEngMap(img)
M_up, Tb_up = cumMinEngHor.cumMinEngHor(e_up)
image_up, cost_up = rmHorSeam.rmHorSeam(img, M_up, Tb_up)
cmp_list = [Tr[i, j - 1] + cost_left, Tr[i - 1, j] + cost_up]
min_val = min(cmp_list)
Tr[i, j] = min_val
min_index = cmp_list.index(min_val)
T[i, j] = min_index
if (min_index == 0):
image_list[i][j] = np.copy(image_left)
else:
image_list[i][j] = np.copy(image_up)
Ic = image_list[r - 1][c - 1]
#Saving the last frame
im = Image.fromarray(Ic)
im.save('Scenary_final.jpg')
#creating the GIF
gif_list = [[] for j in range(nr + nc + 1)]
i = nr + nc
carved_list = backtrack(T, nr, nc, image_list, gif_list, i)
imageio.mimsave('Scenary.gif', carved_list, duration=0.2)
return Ic, Tr
plt.figure(1) plt.imshow(I) # auto-refined mask x1 = ind[1] y1 = ind[0] x2 = ind[3] y2 = ind[2] mask, h, w = Domask(I, x1, x2, y1, y2) I_label = label(I, mask, x1, x2, y1, y2) plt.figure(2) plt.imshow(I_label) # object removal with weighted seam carving nr = 0 nc = ind[3] - ind[1] [Ic, T] = carv(I, nr, nc, I_label) fig2 = plt.figure(3) plt.imshow(Ic) # seam insertion Ic_n = Ic summ = int(nc / 2) I_label_insert = np.ones(np.shape(Ic_n[:, :, 0])) a = genEngMap(Ic_n, I_label_insert) [Mx, Tbx] = cumMinEngVer(a) [Ix, Ex] = insertVerSeam(Ic_n, Mx, Tbx, summ) plt.figure(4) plt.imshow(Ix) plt.show()