def judge_area_similar(foreground_area,
background_area,
similar_threshold=0.5):
"""
判断两个区域是否相似，这有利于poisson编辑的效果，这只是固定位置判断，输入的都是两个mask区域
1.判断直方图
2. 判断
:param foreground_area:
:param background_area:
:param similar_threshold:有百分之多少的相似 默认 0.5
:return: 返回一个bool数,表示该选择的预期ok否
"""
# 直方图
similar_score = myCalImage.calc_similar(foreground_area, background_area)
print('直方图相似程度:', similar_score)
if similar_score > 0.5 * 100:
return True
else:
return False
def splicing_tamper_one_image(input_foreground,
input_background,
mask,
similar_judge=True):
"""
在输入这个函数之前，输入的两个参数需要以及经过挑选了的,需要import poisson融合的代码
:param foreground:
:param background:
:return: 返回两个参数：直接篡改图, poisson融合篡改图, GT
"""
I = input_foreground
I1 = I
# mask 是 01 蒙版
I1[:, :, 0] = np.array(I[:, :, 0] * mask)
I1[:, :, 1] = np.array(I[:, :, 1] * mask)
I1[:, :, 2] = np.array(I[:, :, 2] * mask)
# differece_8是background的edge
difference_8 = mask_to_outeedge(mask)
difference_8_dilation = dilation.binary_dilation(difference_8,
np.ones((3, 3)))
difference_8_dilation = np.where(difference_8_dilation == True, 1, 0)
double_edge_candidate = difference_8_dilation + mask
double_edge = np.where(double_edge_candidate == 2, 1, 0)
ground_truth = np.where(double_edge == 1, 255, 0) + np.where(
difference_8 == 1, 100, 0) + np.where(mask == 1, 50,
0) # 所以内侧边缘就是100的灰度值
b1 = input_background
background = Image.fromarray(b1, 'RGB')
foreground = Image.fromarray(I1, 'RGB').convert('RGBA')
datas = foreground.getdata()
newData = []
for item in datas:
if item[0] == 0 and item[1] == 0 and item[2] == 0:
newData.append((0, 0, 0, 0))
else:
newData.append(item)
foreground.putdata(newData)
background = background.resize((foreground.size[0], foreground.size[1]),
Image.ANTIALIAS)
# input_background = np.resize(input_background,(foreground.size[1], foreground.size[0],3))
background_area = np.array(background)
try:
if similar_judge:
foreground_area = I1
background_area[:, :,
0] = np.array(background_area[:, :, 0] * mask)
background_area[:, :,
1] = np.array(background_area[:, :, 1] * mask)
background_area[:, :,
2] = np.array(background_area[:, :, 2] * mask)
foreground_area = Image.fromarray(foreground_area)
background_area = Image.fromarray(background_area)
if myCalImage.calc_similar(foreground_area, background_area) > 0.2:
pass
else:
return False, False, False
except Exception as e:
print(e)
traceback.print_exc()
try:
mask = Image.fromarray(mask)
except Exception as e:
print('mask to Image error', e)
# 在这里的时候，mask foreground background 尺寸都是一致的了，poisson融合时，offset置为0
try:
poisson_foreground = cv2.cvtColor(
np.asarray(foreground.convert('RGB')), cv2.COLOR_RGB2BGR)
poisson_background = cv2.cvtColor(np.asarray(background),
cv2.COLOR_RGB2BGR)
poisson_mask = np.asarray(mask)
poisson_mask = np.where(poisson_mask == 1, 255, 0)
poisson_fusion_image = poisson_image_editing.poisson_fusion(
poisson_foreground, poisson_background, poisson_mask)
poisson_fusion_image = Image.fromarray(
cv2.cvtColor(poisson_fusion_image, cv2.COLOR_BGR2RGB))
background.paste(foreground, (0, 0), mask=foreground.split()[3])
return background, poisson_fusion_image, ground_truth
except Exception as e:
traceback.print_exc()