CV基础 —— 泊松融合(代码)

图像处理

from matplotlib import pylab as plt
import numpy as np
from PIL import Image
import cv2

image_src = Image.open('./test2_src.png').convert('RGB')
image_target = Image.open('./test2_target.png').convert('RGB')
im_src = np.array(image_src)
im_target = np.array(image_target)

plt.subplot(1,2,1)
plt.imshow(im_src)
plt.subplot(1,2,2)
plt.imshow(im_target)

<matplotlib.image.AxesImage at 0x7f68f7c34b38>

im_src.shape, im_target.shape

((120, 180, 3), (356, 418, 3))

# 转换后的位置，0点算起，非中心点
position = (150, 150)

# 区域mask
omega = np.zeros((im_target.shape[0], im_target.shape[1]), dtype=np.bool)
# 删除最外面一圈点，防止边缘计算梯度错误
omega[position[0] + 1: position[0] + im_src.shape[0] -1, position[1] + 1: position[1] + im_src.shape[1] - 1] = 1
mask_target = im_target.copy()
mask_target[position[0]: position[0] + im_src.shape[0], position[1]: position[1] + im_src.shape[1], :] = im_src

plt.subplot(1,2,1)
plt.imshow(omega)
plt.subplot(1,2,2)
plt.imshow(mask_target)

<matplotlib.image.AxesImage at 0x7f68f639e8d0>

def in_omega(omega, position):
    x, y = position
    if omega[x, y] == True:
        return True
    else:
        return False

def get_neighbors(position):
    x, y = position
    return (x - 1, y), (x + 1, y), (x, y - 1), (x, y + 1)
def laplace(X, position):
    x, y = position
    res = -4 * X[x, y]
    neighbors = get_neighbors(position)
    for near_x, near_y in neighbors:
        res += X[near_x, near_y]
    return res

maps = {}
index = 0
for i in range(im_target.shape[0]):
    for j in range(im_target.shape[1]):
        maps[(i, j)] = index
        index += 1

# 稀疏矩阵来保存系数矩阵
import scipy.sparse as sp

A = sp.lil_matrix((len(maps), len(maps)))
for i in range(im_target.shape[0]):
    for j in range(im_target.shape[1]):
        point_c = maps[(i, j)]
        if in_omega(omega, (i, j)) is not True:
            # 不在范围内
            A[point_c, point_c] = 1
        else:
            # 离散拉普拉斯算子
            A[point_c, point_c] = -4
            c_neighbors = get_neighbors((i, j))
            for neighbor in c_neighbors:
                near_x, near_y = neighbor
                point_near = maps[(near_x, near_y)]
                A[point_c, point_near] = 1

# 与A过程相同，可以合并
b = np.zeros((len(maps), 3))
for i in range(im_target.shape[0]):
    for j in range(im_target.shape[1]):
        point_c = maps[(i, j)]
        if in_omega(omega, (i, j)) is not True:
            # 不在范围内
            b[point_c, :] = im_target[i, j, :]
        else:
            b[point_c, :] = laplace(mask_target, (i, j))

import scipy.sparse.linalg as spl

# 解方程组Af = b
f = spl.spsolve(A.tocsc(), b)

final_target = np.zeros(im_target.shape, dtype=np.uint8)

final_target[:,:,0] = f[:,0].reshape(omega.shape)
final_target[:,:,1] = f[:,1].reshape(omega.shape)
final_target[:,:,2] = f[:,2].reshape(omega.shape)

plt.figure(figsize=(20,40), dpi=50)
plt.subplot(1,2,1)
plt.imshow(mask_target)
plt.subplot(1,2,2)
plt.imshow(final_target)

<matplotlib.image.AxesImage at 0x7f689a5e7f60>

np.where(final_target == 0)

(array([153, 157, 157, 165, 172, 221, 229, 239, 242, 243, 248, 248, 251,
        255, 256, 257, 258, 258, 258, 258, 259, 260, 260, 261, 262, 262,
        262, 262, 262, 263, 263, 263, 263, 264, 265, 265, 265, 265, 266,
        266, 267, 267, 267, 267, 267, 268, 268, 268]),
 array([326, 317, 324, 322, 296, 168, 158, 162, 160, 167, 159, 160, 161,
        168, 160, 159, 155, 157, 164, 166, 163, 156, 168, 171, 156, 163,
        164, 166, 174, 154, 159, 164, 165, 152, 163, 170, 173, 237, 169,
        214, 178, 179, 199, 235, 239, 177, 230, 234]),
 array([0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0]))

np.where(f > 255)

(array([ 63439,  63444,  63445,  63446,  63860,  63861,  63862,  64276,
         64280,  64281,  64282,  65535,  65943,  65948,  65950,  66370,
         67196,  67626,  68035,  69292,  70502,  71751,  71756,  72171,
         72178,  72595,  73014,  82187,  82188,  82189,  82605,  82605,
         82605,  82606,  82606,  82607,  82607,  83023,  83023,  83023,
         83024,  83024,  83025,  83025,  83026,  83026,  83441,  83442,
         83442,  83443,  83443,  83444,  83444,  83859,  83860,  83860,
         83861,  83861,  83862,  83862,  84277,  84277,  84278,  84278,
         84279,  84279,  84280,  84280,  84695,  84695,  84696,  84696,
         84697,  84697,  84698,  84698,  85113,  85113,  85114,  85114,
         85115,  85115,  85116,  85116,  85532,  85532,  85533,  85533,
         86404,  88767,  90864,  90875,  91277,  91289,  91294,  91711,
         91713,  92130,  92131,  92545,  92546,  92954,  92959,  92961,
         93380,  93381,  93799,  93801,  94212,  94632,  94737,  95037,
         95041,  95053,  95151,  95455,  95880,  96712,  96715,  97135,
         97550,  97557,  97967,  98387,  98390,  99642,  99644,  99645,
         99646, 100061, 100062, 100063, 100064, 100480, 100484, 101316,
        101318, 101319, 101728, 101741, 102147, 102154, 102565, 102566,
        102988, 103401, 103408, 103412, 103822, 103823, 103824, 103833,
        104234, 104235, 104237, 104238, 104242, 104653, 104658, 104661,
        104662, 104664, 105069, 105071, 105077, 105079, 105080, 105491,
        105492, 105493, 105496, 105498, 105499, 105912, 105913, 106328,
        106331, 106334, 106746, 106748, 106758, 107164, 107168, 107172,
        107173, 107174, 107178, 107582, 107583, 107585, 107586, 107592,
        107999, 108000, 108001, 108002, 108005, 108006, 108007, 108008,
        108010, 108416, 108417, 108420, 108421, 108422, 108424, 108425,
        108426, 108427, 108430, 108432, 108434, 108835, 108836, 108840,
        108841, 108842, 108843, 108844, 108845, 108848, 108849, 108853,
        109254, 109255, 109258, 109259, 109261, 109265, 109267, 109269,
        109329, 109667, 109672, 109673, 109674, 109676, 109679, 109680,
        109682, 109689, 109690, 110088, 110093, 110094, 110095, 110096,
        110098, 110099, 110104, 110504, 110508, 110515, 110517, 110518,
        110521, 110523, 110524, 110527, 110598, 110932, 110933, 110934,
        110940, 110941, 110943, 110974, 110978, 110991, 111007, 111346,
        111351, 111352, 111355, 111357, 111361, 111392, 111393, 111402,
        111410, 111420, 111426, 111427, 111761, 111767, 111768, 111769,
        111771, 111772, 111773, 111775, 111777, 111783, 111784, 111785,
        111805, 111826, 111830, 111833, 111834, 111841, 111842, 111845,
        111847, 111851, 112179, 112185, 112186, 112187, 112189, 112193,
        112195, 112201, 112203, 112223, 112249, 112254, 112258, 112260,
        112261, 112262, 112264, 112265, 112266, 112267, 112268, 112271,
        112275]),
 array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 1, 1, 1, 0, 1, 2, 0, 1, 0, 1, 0, 1, 2, 0, 1, 0, 1,
        0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1,
        0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
        0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]))

# 计算出的f存在大于255的像素，转换成unit8时变成了0
f[f > 255] = 255
f[f < 0] = 0
final_target_corr = np.zeros(im_target.shape, dtype=np.uint8)
final_target_corr[:,:,0] = f[:,0].reshape(omega.shape)
final_target_corr[:,:,1] = f[:,1].reshape(omega.shape)
final_target_corr[:,:,2] = f[:,2].reshape(omega.shape)

plt.figure(figsize=(20,40), dpi=50)
plt.subplot(1,2,1)
plt.imshow(mask_target)
plt.subplot(1,2,2)
plt.imshow(final_target_corr)

<matplotlib.image.AxesImage at 0x7f689a1d3668>