python生成马赛克图片
2020-10-27 本文已影响0人
Aedda
下方注释的是爬取图片的爬虫
爬取的图片集放在同级目录image文件夹下,每一个像素点都是一个图片
import os
import sys
import time
import math
import numpy as np
from PIL import Image, ImageOps
from multiprocessing import Pool
from colorsys import rgb_to_hsv, hsv_to_rgb
from typing import List, Tuple, Union
class mosaic(object):
"""定义计算图片的平均hsv值
"""
def __init__(self, IN_DIR: str, OUT_DIR: str, SLICE_SIZE: int, REPATE: int,OUT_SIZE: int) -> None:
self.IN_DIR = IN_DIR # 原始的图像素材所在文件夹
self.OUT_DIR = OUT_DIR # 输出素材的文件夹, 这些都是计算过hsv和经过resize之后的图像
self.SLICE_SIZE = SLICE_SIZE # 图像放缩后的大小
self.REPATE = REPATE # 同一张图片可以重复使用的次数
self.OUT_SIZE = OUT_SIZE # 最终图片输出的大小
def resize_pic(self, in_name: str, size: int) -> Image:
"""转换图像大小
"""
img = Image.open(in_name)
img = ImageOps.fit(img, (size, size), Image.ANTIALIAS)
return img
def get_avg_color(self, img: Image) -> Tuple[float, float, float]:
"""计算图像的平均hsv
"""
width, height = img.size
pixels = img.load()
if type(pixels) is not int:
data = [] # 存储图像像素的值
for x in range(width):
for y in range(height):
cpixel = pixels[x, y] # 获得每一个像素的值
data.append(cpixel)
h = 0
s = 0
v = 0
count = 0
for x in range(len(data)):
r = data[x][0]
g = data[x][1]
b = data[x][2] # 得到一个点的GRB三色
count += 1
hsv = rgb_to_hsv(r / 255.0, g / 255.0, b / 255.0)
h += hsv[0]
s += hsv[1]
v += hsv[2]
hAvg = round(h / count, 3)
sAvg = round(s / count, 3)
vAvg = round(v / count, 3)
if count > 0: # 像素点的个数大于0
return (hAvg, sAvg, vAvg)
else:
raise IOError("读取图片数据失败")
else:
raise IOError("PIL 读取图片数据失败")
class create_image_db(mosaic):
"""创建所需要的数据
"""
def __init__(self, IN_DIR: str, OUT_DIR: str, SLICE_SIZE: int, REPATE: int,
OUT_SIZE: int) -> None:
super(create_image_db, self).__init__(IN_DIR, OUT_DIR, SLICE_SIZE,
REPATE, OUT_SIZE)
def make_dir(self) -> None:
os.makedirs(os.path.dirname(self.OUT_DIR), exist_ok=True) # 没有就创建文件夹
def get_image_paths(self) -> List[str]:
"""获取文件夹内图像的地址
"""
paths = []
suffixs = ['png', 'jpg']
for file_ in os.listdir(self.IN_DIR):
# suffix = file_.split('.')[-1] # 获得文件后缀
# if suffix in suffixs: # 通过后缀判断是否是图片
if file_.endswith('jpg'):
paths.append(os.path.join(self.IN_DIR,file_)) # 添加图像路径
else:
print("非图片:%s" % file_)
if len(paths) > 0:
print("一共找到了%s" % len(paths) + "张图片")
else:
raise IOError("未找到任何图片")
return paths
def convert_image(self, path):
"""转换图像大小, 同时计算一个图像的平均hsv值.
"""
try:
img = self.resize_pic(path, self.SLICE_SIZE)
color = self.get_avg_color(img)
img.save(str(self.OUT_DIR) + str(color) + ".png")
except:
pass
def convert_all_images(self) -> None:
"""将所有图像进行转换
"""
self.make_dir()
paths = self.get_image_paths()
print("正在生成马赛克块...")
pool = Pool() # 多进程处理
pool.map(self.convert_image, paths) # 对已有的图像进行处理, 转换为对应的色块
pool.close()
pool.join()
class create_mosaic(mosaic):
"""创建马赛克图片
"""
def __init__(self, IN_DIR: str, OUT_DIR: str, SLICE_SIZE: int, REPATE: int,
OUT_SIZE: int) -> None:
super(create_mosaic, self).__init__(IN_DIR, OUT_DIR, SLICE_SIZE, REPATE,
OUT_SIZE)
def read_img_db(self) -> List[List[Union[float, int]]]:
"""读取所有的图片
"""
img_db = [] # 存储color_list
for file_ in os.listdir(self.OUT_DIR):
if file_ == 'None.png':
pass
else:
file_ = file_.split('.png')[0] # 获得文件名
file_ = file_[1:-1].split(',') # 获得hsv三个值
file_ = [float(i) for i in file_]
file_.append(0) # 最后一位计算图像使用次数
img_db.append(file_)
return img_db
def find_closiest(self, color: Tuple[float, float, float],
list_colors: List[List[Union[float, int]]]) -> str:
"""寻找与像素块颜色最接近的图像
"""
FAR = 10000000
for cur_color in list_colors: # list_color是图像库中所以图像的平均hsv颜色
n_diff = np.sum((color - np.absolute(cur_color[:3]))**2)
if cur_color[3] <= self.REPATE: # 同一个图片使用次数不能太多
if n_diff < FAR: # 修改最接近的颜色
FAR = n_diff
cur_closer = cur_color
cur_closer[3] += 1
return "({}, {}, {})".format(cur_closer[0], cur_closer[1],
cur_closer[2]) # 返回hsv颜色
def make_puzzle(self, img: str) -> bool:
"""制作拼图
"""
img = self.resize_pic(img, self.OUT_SIZE) # 读取图片并修改大小
color_list = self.read_img_db() # 获取所有的颜色的list
width, height = img.size # 获得图片的宽度和高度
print("Width = {}, Height = {}".format(width, height))
background = Image.new('RGB', img.size,
(255, 255, 255)) # 创建一个空白的背景, 之后向里面填充图片
total_images = math.floor(
(width * height) / (self.SLICE_SIZE * self.SLICE_SIZE)) # 需要多少小图片
now_images = 0 # 用来计算完成度
for y1 in range(0, height, self.SLICE_SIZE):
for x1 in range(0, width, self.SLICE_SIZE):
try:
# 计算当前位置
y2 = y1 + self.SLICE_SIZE
x2 = x1 + self.SLICE_SIZE
# 截取图像的一小块, 并计算平均hsv
new_img = img.crop((x1, y1, x2, y2))
color = self.get_avg_color(new_img)
# 找到最相似颜色的照片
close_img_name = self.find_closiest(color, color_list)
close_img_name = self.OUT_DIR + str(
close_img_name) + '.png' # 图片的地址
paste_img = Image.open(close_img_name)
# 计算完成度
now_images += 1
now_done = math.floor((now_images / total_images) * 100)
r = '\r[{}{}]{}%'.format("#" * now_done,
" " * (100 - now_done), now_done)
sys.stdout.write(r)
sys.stdout.flush()
background.paste(paste_img, (x1, y1))
except IOError:
print('创建马赛克块失败')
# 保持最后的结果
background.save('out_without_background.jpg')
img = Image.blend(background, img, 0.5)
img.save('out_with_background.jpg')
return True
if __name__ == "__main__":
filePath = os.path.dirname(os.path.abspath(__file__)) # 获取当前的路径
start_time = time.time() # 程序开始运行时间, 记录一共运行了多久
# 创建马赛克块, 创建素材库
createdb = create_image_db(IN_DIR=os.path.join(filePath, 'images/'),
OUT_DIR=os.path.join(filePath, 'outputImages/'),
SLICE_SIZE=100,
REPATE=20,
OUT_SIZE=8000)
createdb.convert_all_images()
# 创建拼图 (这里使用绝对路径)
createM = create_mosaic(IN_DIR=os.path.join(filePath, 'images/'),
OUT_DIR=os.path.join(filePath, 'outputImages/'),
SLICE_SIZE=100,
REPATE=20,
OUT_SIZE=8000)
out = createM.make_puzzle(img=os.path.join(filePath, '微信图片_20201027102845.jpg'))
# 打印时间
print("耗时: %s" % (time.time() - start_time))
print("已完成")
# 爬图片
# from urllib import request
# from common import Common
# import datetime
#
# driver = Common.get_driver('chrome')
# driver.get('https://pic.sogou.com/pics?ie=utf8&p=40230504&interV=kKIOkrELjboMmLkEkrsTkKIMkLELjboImLkEk74TkKIMkrELjbkRmLkEmrELjbgRmLkEkLYTLlDpFEPQjb0E0vcGwOVFmVnmMz7lWaIPjf9QxeVPz6R7zOMTKVeRFTbxHUwElKJ6wu9DxqRLzO1H1qR7zOM%3D_923767100&query=%E9%A3%9E%E8%BD%A6')
# end_time = datetime.datetime.now() + datetime.timedelta(seconds=15)
# while True:
# driver.execute_script("window.scrollTo(0,document.body.scrollHeight)")
# time.sleep(0.2)
# if datetime.datetime.now() > end_time:
# time.sleep(3)
# break
# f = 0
# for i in driver.find_elements_by_xpath('//img[@src]'):
# try:
# url = i.get_attribute('src')
# request.urlretrieve(url, f'images\\sougou{f}.jpg')
# print(f)
# f+=1
# except:
# print('出错')