# -*- coding: utf-8 -*-
"""
@Time : 2023/2/6 006 下午 17:49
@Author : Jan
@File : 图片处理.py
"""
""" { 切割图片随机组合再还原 & 滑块验证缺口坐标识别 & 图片无损放大 } """
import cv2
import requests
import numpy as np
from io import BytesIO
from PIL import Image, ImageDraw
def dec(function):
return function()
@dec
class SlideCrack(object):
def __int__(self):
pass
def __call__(self, target: [str, bytes], slide: [str, bytes], out: str = 'out'):
"""
:param target: 背景图片path
:param slide: 缺口图片path
:param out: 输出图片path
"""
self.out = out + '.png'
# 1、本地文件路径
if isinstance(target, str):
self.bg = target
self.bg_img = cv2.imread(self.bg)
if isinstance(slide, str):
self.tp = slide
self.tp_img = cv2.imread(self.tp)
# 2、二进制 ---> numpy数组
if isinstance(target, bytes):
self.bg_img = self.bytes_to_np(target)
if isinstance(slide, bytes):
self.tp_img = self.bytes_to_np(slide)
return self.discern()
# 二进制 ---> numpy数组
@classmethod
def bytes_to_np(self, byte: bytes):
data = np.asarray(bytearray(byte), dtype="uint8")
image = cv2.imdecode(data, cv2.IMREAD_COLOR)
return image
# 二进制 ---> PIL
@classmethod
def bytes_to_pil(self, byte: bytes):
img = Image.open(BytesIO(byte))
return img
# numpy数组 ---> 二进制
@classmethod
def np_to_byte(self, image) -> bytes:
res = cv2.imencode('.png', image)[1].tobytes()
return res
# numpy数组 ---> 二进制 ---> PIL
@classmethod
def np_to_pil(self, image):
data = self.np_to_byte(image)
img = self.bytes_to_pil(data)
return img
# PIL ---> 二进制
@classmethod
def pil_to_byte(self, img) -> bytes:
bytesIO = BytesIO() # 注意：针对每个图片都要新建，这里并不代表赋值，感觉像是新建一个实例化对象
img.save(bytesIO, format='PNG') # format='JPEG'，实际未保存
res = bytesIO.getvalue()
return res
# PIL ---> 二进制 ---> numpy数组
@classmethod
def pil_to_np(self, img):
data = self.pil_to_byte(img)
image = self.bytes_to_np(data)
return image
@classmethod
def show(self, im: [any, bytes]):
if isinstance(im, bytes):
img = self.bytes_to_pil(im)
else:
img = self.np_to_pil(im)
img.show()
# 不推荐，会暂停占用程序运行时间
# cv2.imshow('Show', im)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
@staticmethod
def clear_white(img):
rows, cols, channel = img.shape
min_x = min_y = 255
max_x = max_y = 0
for x in range(1, rows):
for y in range(1, cols):
t = set(img[x, y])
if len(t) >= 2:
if x <= min_x:
min_x = x
elif x >= max_x:
max_x = x
if y <= min_y:
min_y = y
elif y >= max_y:
max_y = y
return img[min_x:max_x, min_y: max_y]
def template_match(self, target, slide):
th, tw = slide.shape[:2]
result = cv2.matchTemplate(target, slide, cv2.TM_CCOEFF_NORMED)
# 寻找矩阵(一维数组当作向量,用Mat定义) 中最小值和最大值的位置
min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(result)
tl = max_loc
br = (tl[0] + tw, tl[1] + th)
return tl, br
# 绘制矩形边框，将匹配区域标注出来
@classmethod
def image_draw_save(self, target: [any, bytes], start, end, offset=0):
if isinstance(target, bytes):
target = self.bytes_to_np(target)
# target：目标图像
# start：矩形定点
# end：矩形的终点
# (0,0,255)：矩形边框颜色
# 1：矩形边框线粗细
start = (lambda x: (x[0] - offset, x[1] - offset))(start)
end = (end[0] + offset, end[1] + offset)
# # 方式一
# cv2.rectangle(target, start, end, (0, 0, 255), 1)
# a = 'x, y' # 类别名称
# b = start # 置信度
# font = cv2.FONT_HERSHEY_SIMPLEX # 定义字体
# imgzi = cv2.putText(target, '{} = {}'.format(a, b), (start[0], start[1] - 15), font, 1, (0, 0, 255), 2)
# 图像， 文字内容， 坐标(右上角坐标) 字体，大小， 颜色， 字体厚度
# 显示，保存
# cv2.imwrite(self.out, target)
# self.show(imgzi)
# 方式二，使用PIL
img = self.np_to_pil(target)
a = ImageDraw.ImageDraw(img)
a.rectangle((start, end), fill=None, outline='red', width=1)
# 显示，保存
# img.save(self.out)
img.show()
def discern(self):
# 清除图片的空白区域
self.bg_img = self.clear_white(self.bg_img)
self.tp_img = self.clear_white(self.tp_img)
# 转换图片为灰色
bg_gray = cv2.cvtColor(self.bg_img, cv2.COLOR_RGB2GRAY)
tp_gray = cv2.cvtColor(self.tp_img, cv2.COLOR_RGB2GRAY)
# 识别图片边缘
bg_edge = cv2.Canny(bg_gray, 100, 200)
tp_edge = cv2.Canny(tp_gray, 100, 200)
# 转换图片格式
bg_pic = cv2.cvtColor(bg_edge, cv2.COLOR_GRAY2RGB)
tp_pic = cv2.cvtColor(tp_edge, cv2.COLOR_GRAY2RGB)
# 获取缺口起止坐标
start, end = self.template_match(bg_pic, tp_pic)
self.image_draw_save(self.bg_img, start, end, offset=2) # 可调整线框偏移量
print("point:", start)
return start
# 图片无损放大
def retinapad_picture(src: [str, bytes], multiple=2):
def judg(src: [str, bytes]):
img = None
# 1、本地文件路径
if isinstance(src, str):
img = cv2.imread(src)
# 2、二进制 ---> numpy数组
if isinstance(src, bytes):
img = SlideCrack.bytes_to_np(src)
return img
img = judg(src)
height, width = img.shape[:2]
image = SlideCrack.np_to_pil(img)
li = []
for y in range(height): # 循环读取图片的每一个像素点的RGBA值 获取图片的Y轴有多少像素 也相当于长度
tmp = []
for x in range(width): # 获取图片的X轴有多少像素 也相当于宽度
# image.getpixel((a, aa)) 用来获取图片某位置的RGBA像素值
tmp = tmp + [image.getpixel((x, y))] * multiple # 读取某坐标的像素值并将元组为列表进行存储 Multiple是倍数
li += tmp * multiple # Multiple是倍数 如果是2倍 则生成两个同样的颜色文件 在后期进行单行输出多次 确保以像素点进行放大
# 图像缩放 要将原图进行翻倍放大 然后在原图的基础上进行绘图
result = cv2.resize(img, (width * multiple, height * multiple))
IMAGE = SlideCrack.np_to_pil(result)
# 转化为RGBA
RGBA_IMG = IMAGE.convert("RGBA")
RGBA_IMG.putdata(li) # 写入图片
RGBA_IMG.show()
# 切割图片，按几行几列分割
def get_split_img(img, row=2, col=2) -> dict:
width, height = img.size
x, y = width // col, height // row
img = img.resize((x * col, y * row)) # 非整数倍切割时对其缩小至整倍数
width, height = img.size
ims = {}
key = 0
for j in range(0, height, y):
for i in range(0, width, x):
# image = SlideCrack.pil_to_np(img)
# im = image[j:j + y, i:i + x] # 使用 OpenCV 模块，裁剪坐标为[y0:y1, x0:x1]
# im = SlideCrack.np_to_pil(im)
im = img.crop((i, j, i + x, j + y)) # 使用 Pillow 模块，坐标(left, upper, right, lower)
ims.update({key: im})
key += 1
return ims
def get_random_img(ims, img, row=2):
pass
x = y = 0
col = len(ims) // row
width, height = ims[0].size
result = Image.new(ims[0].mode, (width * col, height * row))
for im in ims:
result.paste(im, box=(x * width, y))
x += 1
if x % col == 0:
y += height
x -= col
result = result.resize(img.size) # 此处缩放回原来大小会有误差导致还原后失真，不过有真实拼凑的间隙感，可选项
result.show()
return result
# 上下拼接
# width, height = ims[0].size
# result = Image.new(ims[0].mode, (width, height * len(ims)))
# for j, im in enumerate(ims):
# result.paste(im, box=(0, j * height))
# result.show()
# 左右拼接
# width, height = ims[0].size
# result = Image.new(ims[0].mode, (width * len(ims), height))
# for j, im in enumerate(ims):
# result.paste(im, box=(j * width, 0))
# result.show()
def get_recover_img(img, serilize: [list, tuple], row=2):
col = len(serilize) // row
ims = get_split_img(img, row=row, col=col)
index = sorted(range(len(serilize)), key=lambda x: serilize[x]) # 还原初始位置索引
ims = [ims[i] for i in index]
result = get_random_img(ims, img, row)
return result
def random_combined(ims, serilize: list = None):
# 打乱列表顺序，随机排序
lr = [i for i in range(len(ims))]
serilize = np.random.choice(lr, len(lr), replace=False) if not serilize else serilize
ims = [ims[i] for i in serilize]
return ims, list(serilize)
def main(img, row=2, col=2, serilize: list = None):
# 切割图片
ims = get_split_img(img, row=row, col=col)
# 打乱后的图片
ims, serilize = random_combined(ims, serilize)
# # 1、继续打乱切割后指定的位置
# def other(im, row, col, serilize: list = None):
# ims4 = get_split_img(im, row=row, col=col)
# ims4, serilize1 = random_combined(ims4, serilize=serilize)
# out = get_random_img(ims4, img=img, row=row)
# return out
#
# # ims[serilize.index(4)] = other(ims[serilize.index(4)], row=3, col=3, serilize=None) # 原图的中间 serilize.index(4)
# ims[4] = other(ims[4], row=3, col=3, serilize=None) # 打乱后的中间 4
# 2、放个缩小图
ims[(row * col) // 2] = img.resize(ims[0].size) # 这样无告警提示
# 拼接打乱后的图片
out = get_random_img(ims, img, row=row)
# out.save('C:/Users/11390/Desktop/22.PNG')
# 还原
result = get_recover_img(out, serilize=serilize, row=row)
if __name__ == '__main__':
pass
# 切割图片并随机排列再还原
url = 'https://uploadfiles.nowcoder.com/images/20230206/993219759_1675676787618/156005C5BAF40FF51A327F1C34F2975B'
data = requests.get(url).content
# SlideCrack.show(data)
img = SlideCrack.bytes_to_pil(data)
img = img.convert('RGBA') # (此处有个小坑) 需要图像数据的预处理，从'P','RGBA','RGB'等多种 mode 的图像转换
# 切割行列可以是任意整数，按宽高比例整倍数最佳
row = 7
col = 7
main(img, row=row, col=col)
# # 放大图片
# k = 8
# url = 'http://ydgw.yundasys.com:31620/upload/image/20220826/aa710e4b01c732cb2050ba5db845af76.png'
# data = requests.get(url).content
# SlideCrack.show(data)
# src = SlideCrack.bytes_to_np(data)
# y, x = src.shape[:2]
# res = cv2.resize(src, (x * k, y * k))
# SlideCrack.show(SlideCrack.np_to_byte(res))
#
# # 无损放大
# retinapad_picture(data, k)
投递旷视等公司10个岗位 >
拼多多集团-PDD工作强度 274人发布
