95 lines
3.4 KiB
Python
95 lines
3.4 KiB
Python
# -*- coding:utf-8 -*-
|
|
# @Author len
|
|
# @Create 2023/12/25 14:55
|
|
|
|
import cv2
|
|
import numpy as np
|
|
import random
|
|
import os
|
|
|
|
def coordinates2yolo(xmin,ymin,xmax,ymax):
|
|
img_w, img_h = 800, 480
|
|
# 保留6位小数
|
|
x = round((xmin+xmax)/(2.0*img_w),6)
|
|
y = round((ymin+ymax)/(2.0*img_h),6)
|
|
w1 = round((xmax-xmin)/(1.0*img_w),6)
|
|
h1 = round((ymax-ymin)/(1.0*img_h),6)
|
|
# print( x,y,w1,h1)
|
|
return x,y,w1,h1
|
|
|
|
def process_image(bg_path, logo_path, output_path):
|
|
|
|
background = cv2.imread(bg_path)
|
|
logo = cv2.imread(logo_path, -1) # -1 表示尝试读取透明度通道
|
|
flag = logo_path.split('\\')[-1][0]
|
|
|
|
if logo is None or background is None:
|
|
print(f"无法读取图像，跳过 {bg_path} 和 {logo_path}")
|
|
return
|
|
|
|
# 检查标志图像是否有透明度通道
|
|
if logo.shape[2] == 4:
|
|
alpha_channel = logo[:, :, 3]
|
|
else:
|
|
alpha_channel = np.ones(logo.shape[:2], dtype=logo.dtype) * 255
|
|
|
|
# 确保标志图像小于背景图像
|
|
if logo.shape[1] > background.shape[1] or logo.shape[0] > background.shape[0]:
|
|
scale_factor = min(background.shape[1] / logo.shape[1], background.shape[0] / logo.shape[0]) * 0.9
|
|
else:
|
|
scale_factor = random.uniform(0.5, 1.5)
|
|
|
|
width = int(logo.shape[1] * scale_factor)#图像长宽
|
|
height = int(logo.shape[0] * scale_factor)
|
|
logo = cv2.resize(logo, (width, height))
|
|
alpha_channel = cv2.resize(alpha_channel, (width, height))
|
|
|
|
# 确保随机位置的计算是有效的
|
|
x_offset = random.randint(0, max(0, background.shape[1] - width))
|
|
y_offset = random.randint(0, max(0, background.shape[0] - height))
|
|
|
|
# 计算标志图片的右下角坐标
|
|
x_end = x_offset + width+random.randint(0, 5)
|
|
y_end = y_offset + height+random.randint(0, 5)
|
|
result1 = coordinates2yolo(x_offset-random.randint(0, 5), y_offset-random.randint(0, 5), x_end, y_end)
|
|
|
|
result = (flag + " " + str(result1[0]) + " " + str(result1[1]) + " " + str(result1[2]) + " " + str(result1[3]))
|
|
path_txt = output_path.split('.')[0] + '.txt'
|
|
# print(result)
|
|
# return
|
|
with open(path_txt, 'w') as f:
|
|
f.write(result)
|
|
|
|
# 合成图像
|
|
for c in range(0, 3):
|
|
background[y_offset:y_offset+height, x_offset:x_offset+width, c] = \
|
|
logo[:, :, c] * (alpha_channel / 255.0) + \
|
|
background[y_offset:y_offset+height, x_offset:x_offset+width, c] * (1.0 - alpha_channel / 255.0)
|
|
|
|
# 保存结果到文件
|
|
cv2.imwrite(output_path, background)
|
|
|
|
|
|
# 文件夹路径
|
|
backgrounds_path = r'D:\1225\back' # 替换为你的背景图片文件夹路径
|
|
logos_path = r'D:\1225\sign' # 替换为你的标志图片文件夹路径
|
|
output_path = r'D:\1225\add' # 替换为你的输出文件夹路径
|
|
|
|
# 确保输出路径存在
|
|
if not os.path.exists(output_path):
|
|
os.makedirs(output_path)
|
|
|
|
backgrounds = [f for f in os.listdir(backgrounds_path) if f.endswith(('.png', '.jpg', '.jpeg'))]
|
|
logos = [f for f in os.listdir(logos_path) if f.endswith(('.png', '.jpg', '.jpeg'))]
|
|
|
|
# 为每个背景图片和每个标志图片组合
|
|
for bg_file in backgrounds:
|
|
for logo_file in logos:
|
|
for i in range(8,10):
|
|
bg_full_path = os.path.join(backgrounds_path, bg_file)
|
|
logo_full_path = os.path.join(logos_path, logo_file)
|
|
output_full_path = os.path.join(output_path, f'combined{i}_{bg_file[:-4]}_{logo_file}')
|
|
process_image(bg_full_path, logo_full_path, output_full_path)
|
|
|
|
print("批量处理完成！")
|