摘要：这篇文章介绍微信小程序“大家来找茬”怎么使用程序自动“找茬”，使用到的工具主要是Python3和adb工具。
作者：yooongchun
微信公众号： yooongchun小屋
腾讯官方出了一个小程序的找茬游戏，如下示意：
很多时候“眼疾手快”比不过别人，只好寻找一种便捷的玩法：程序自动实现！
这里使用的是Python3
- 第一步：获取手机截图
os.system("adb.exe exec-out screencap -p >screenshot.png")上面的命令获得的截图在windows系统上会出错，这是由于windows默认使用的换行符为
而Andriod系统使用的是Linux内核，其换行表示为
，在手机端把二进制数据流传输给电脑时，windows会自动把
替换为
因而为了正确显示，还需要一个转换，我们编写Python的转换代码如下：
# 转换图片格式
# adb 工具直接截图保存到电脑的二进制数据流在windows下"
" 会被解析为"
",
# 这是由于Linux系统下和Windows系统下表示的不同造成的，而Andriod使用的是Linux内核
def convert_img(path):
with open(path, "br") as f:
bys = f.read()
bys_ = bys.replace(b"
", b"
") # 二进制流中的"
" 替换为"
"
with open(path, "bw") as f:
f.write(bys_)
- 第二步：图片裁剪
获得的图片有多余的部分，需要进行裁剪，使用Python的opencv库，代码如下：
# 裁剪图片
def crop_image(im, box=(0.20, 0.93, 0.05, 0.95), gap=38, dis=2):
'''
:param path: 图片路径
:param box: 裁剪的参数：比例
:param gap: 中间裁除区域
:param dis: 偏移距离
:return: 返回裁剪出来的区域
'''
h, w = im.shape[0], im.shape[1]
region = im[int(h * box[2]):int(h * box[3]), int(w * box[0]):int(w * box[1])]
rh, rw = region.shape[0], region.shape[1]
region_1 = region[0 + dis: int(rh / 2) - gap + dis, 0: rw]
region_2 = region[rh - int(rh / 2) + gap: rh, 0:rw]
return region_1, region_2, region
- 第三步：图片差异对比
图片差异对比这就很好理解了，把两张图片叠到一起，相减，剩下的就是不同的地方了，当然，这里有几个细节需要注意：原图的截取，上面从手机获取的截图有很多非目标区域，因而我们需要定义截图区域，这就是我们程序中需要给出的box参数：
box=(0.2,0.93,0.05,0.95)
这里，参数依次代表：
开始截取的列=0.2*图片宽，停止截取的列=0.93*图片宽
开始截取的行=0.05*图片高，开始截取的行=0.95*图片高
然后，仔细观察你会发现中间还有一块多余的区域，把上下两张图分开只需要给出中间区域要截除的像素值，这也就是我们程序运行的第二个参数：
gap=38
这里代表把第一次截图得到的图片二分后分别截去38像素的高度。
这时，还有一个问题要注意的是，我们截图参数是根据肉眼分辨设置的，你截图的结果可能并不是严格的目标图片的开始行列，这时，得到的两张图片会存在很小的错位，为了微调这个错位，我们给出程序的第三个参数：
dis=2
这代表两张图片在进行相减作差的时候会微调两行。
好了，得到差异图片后我们来看看效果
哈，五个不同的地方，终于“原形毕露”！
# 查找不同返回差值图
def diff(img1, img2):
diff = (img1 - img2)
# 形态学开运算滤波
kernel = np.ones((5, 5), np.uint8)
opening = cv2.morphologyEx(diff, cv2.MORPH_OPEN, kernel)
return opening这时，你就可以看着这张差异图去“找茬”了。
当然，上面这张丑陋的差异图是不能忍受的，没事，我们接着改进。
找到了差异，如何“优雅”的展示差异呢？我的第一反应就是：在原图上画个圈出来，这样既直观又不失“优雅”。好吧，说干就干！
第一步，使用Opencv库检索差异图的轮廓。这里，值得一提的是在图片的右上角有个小程序的返回图标，这会干扰我们提取轮廓，因而需要先把这个图标去除。查找到轮廓之前需要把图片转换为二值图，然后运用形态学开运算去除噪声，这里涉及程序的第四个参数：滤波核尺寸：
filter_sz=25
最后查找外轮廓并根据轮廓周长保存前n个轮廓，这就是程序里的第五个参数：
num=5
然后检测轮廓的最小外接圆，找到圆心和半径，绘制到原图上，效果如下：
这么样，效果是不是更“优雅”一些了呢！
# 去除右上角的多余区域,即显示小程序返回及分享的灰色区域块
def dispose_region(img):
h, w = img.shape[0], img.shape[1]
img[0:int(0.056 * h), int(0.68 * w):w] = 0
return img
# 查找轮廓中心返回坐标值
def contour_pos(img, num=5, filter_size=5):
'''
:param img: 查找的目标图,需为二值图
:param num: 返回的轮廓数量，如果该值大于轮廓总数，则返回轮廓总数
:return: 返回值为轮廓的最小外接圆的圆心坐标和半径，存放在一个list中
'''
position = [] # 保存返回值
# 计算轮廓
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
binary = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
cv2.namedWindow("binary", cv2.WINDOW_NORMAL)
cv2.imshow("binary", binary)
kernel = np.ones((filter_size, filter_size), np.uint8)
opening = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel) # 开运算
cv2.namedWindow("open", cv2.WINDOW_NORMAL)
cv2.imshow("open", opening)
image, contours, hierarchy = cv2.findContours(np.max(opening) - opening, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# 根据轮廓周长大小决定返回的轮廓
arclen = [cv2.arcLength(contour, True) for contour in contours]
arc = arclen.copy()
arc.sort(reverse=True)
if len(arc) >= num:
thresh = arc[num - 1]
else:
thresh = arc[len(arc) - 1]
for index, contour in enumerate(contours):
if cv2.arcLength(contour, True) < thresh:
continue
(x, y), radius = cv2.minEnclosingCircle(contour)
center = (int(x), int(y))
radius = int(radius)
position.append({"center": center, "radius": radius})
return position
# 在原图上显示
def dip_diff(origin, region, region_1, region_2, dispose_img, position, box, setting_radius=40, gap=38, dis=2):
for pos in position:
center, radius = pos["center"], pos["radius"]
if setting_radius is not None:
radius = setting_radius
cv2.circle(region_2, center, radius, (0, 0, 255), 5)
cv2.namedWindow("region2",cv2.WINDOW_NORMAL)
cv2.imshow("region2",region_2)
gray = cv2.cvtColor(dispose_img, cv2.COLOR_BGR2GRAY)
binary = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
kernel = np.ones((15, 15), np.uint8)
opening = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel) # 开运算
merged = 255 - cv2.merge([opening, opening, opening])
h, w = region_1.shape[0], region_1.shape[1]
region[0:h, 0:w] *= merged
region[0:h, 0:w] += region_1
region[h + gap * 2 - dis:2 * h + gap * 2 - dis, 0:w] = region_2
orih, oriw = origin.shape[0], origin.shape[1]
origin[int(orih * box[2]):int(orih * box[3]), int(oriw * box[0]):int(oriw * box[1])] = region
cv2.namedWindow("show diff", cv2.WINDOW_NORMAL)
cv2.imshow("show diff", origin)
cv2.waitKey(0)
# 自动点击
def auto_click(origin, region_1, box, position, gap=38, dis=2):
h, w = origin.shape[0], origin.shape[1]
rh = region_1.shape[0]
for pos in position:
center, radius = pos["center"], pos["radius"]
x = int(w * box[0] + center[0])
y = int(h * box[2] + rh - dis + 2 * gap + center[1])
os.system("adb.exe shell input tap %d %d" % (x, y))
logging.info("tap:(%d,%d)" % (x, y))
time.sleep(0.05)
最后贴上完整的代码：
"""
大家来找茬微信小程序腾讯官方版 自动找出图片差异
"""
__author__ = "yooongchun"
__email__ = "yooongchun@foxmail.com"
__site__ = "www.yooongchun.com"
import cv2
import numpy as np
import os
import time
import sys
import logging
import threading
logging.basicConfig(level=logging.INFO)
DEBUG = True # 开启debug模式,供调试用，正常使用情况下请不要修改
# 转换图片格式
# adb 工具直接截图保存到电脑的二进制数据流在windows下"
" 会被解析为"
",
# 这是由于Linux系统下和Windows系统下表示的不同造成的，而Andriod使用的是Linux内核
def convert_img(path):
with open(path, "br") as f:
bys = f.read()
bys_ = bys.replace(b"
", b"
") # 二进制流中的"
" 替换为"
"
with open(path, "bw") as f:
f.write(bys_)
# 裁剪图片
def crop_image(im, box=(0.20, 0.93, 0.05, 0.95), gap=38, dis=2):
'''
:param path: 图片路径
:param box: 裁剪的参数：比例
:param gap: 中间裁除区域
:param dis: 偏移距离
:return: 返回裁剪出来的区域
'''
h, w = im.shape[0], im.shape[1]
region = im[int(h * box[2]):int(h * box[3]), int(w * box[0]):int(w * box[1])]
rh, rw = region.shape[0], region.shape[1]
region_1 = region[0 + dis: int(rh / 2) - gap + dis, 0: rw]
region_2 = region[rh - int(rh / 2) + gap: rh, 0:rw]
return region_1, region_2, region
# 查找不同返回差值图
def diff(img1, img2):
diff = (img1 - img2)
# 形态学开运算滤波
kernel = np.ones((5, 5), np.uint8)
opening = cv2.morphologyEx(diff, cv2.MORPH_OPEN, kernel)
return opening
# 去除右上角的多余区域,即显示小程序返回及分享的灰色区域块
def dispose_region(img):
h, w = img.shape[0], img.shape[1]
img[0:int(0.056 * h), int(0.68 * w):w] = 0
return img
# 查找轮廓中心返回坐标值
def contour_pos(img, num=5, filter_size=5):
'''
:param img: 查找的目标图,需为二值图
:param num: 返回的轮廓数量，如果该值大于轮廓总数，则返回轮廓总数
:return: 返回值为轮廓的最小外接圆的圆心坐标和半径，存放在一个list中
'''
position = [] # 保存返回值
# 计算轮廓
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
binary = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
cv2.namedWindow("binary", cv2.WINDOW_NORMAL)
cv2.imshow("binary", binary)
kernel = np.ones((filter_size, filter_size), np.uint8)
opening = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel) # 开运算
cv2.namedWindow("open", cv2.WINDOW_NORMAL)
cv2.imshow("open", opening)
image, contours, hierarchy = cv2.findContours(np.max(opening) - opening, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# 根据轮廓周长大小决定返回的轮廓
arclen = [cv2.arcLength(contour, True) for contour in contours]
arc = arclen.copy()
arc.sort(reverse=True)
if len(arc) >= num:
thresh = arc[num - 1]
else:
thresh = arc[len(arc) - 1]
for index, contour in enumerate(contours):
if cv2.arcLength(contour, True) < thresh:
continue
(x, y), radius = cv2.minEnclosingCircle(contour)
center = (int(x), int(y))
radius = int(radius)
position.append({"center": center, "radius": radius})
return position
# 在原图上显示
def dip_diff(origin, region, region_1, region_2, dispose_img, position, box, setting_radius=40, gap=38, dis=2):
for pos in position:
center, radius = pos["center"], pos["radius"]
if setting_radius is not None:
radius = setting_radius
cv2.circle(region_2, center, radius, (0, 0, 255), 5)
cv2.namedWindow("region2",cv2.WINDOW_NORMAL)
cv2.imshow("region2",region_2)
gray = cv2.cvtColor(dispose_img, cv2.COLOR_BGR2GRAY)
binary = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
kernel = np.ones((15, 15), np.uint8)
opening = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel) # 开运算
merged = 255 - cv2.merge([opening, opening, opening])
h, w = region_1.shape[0], region_1.shape[1]
region[0:h, 0:w] *= merged
region[0:h, 0:w] += region_1
region[h + gap * 2 - dis:2 * h + gap * 2 - dis, 0:w] = region_2
orih, oriw = origin.shape[0], origin.shape[1]
origin[int(orih * box[2]):int(orih * box[3]), int(oriw * box[0]):int(oriw * box[1])] = region
cv2.namedWindow("show diff", cv2.WINDOW_NORMAL)
cv2.imshow("show diff", origin)
cv2.waitKey(0)
# 在原图上绘制圆
def draw_circle(origin, region_1, position, box, gap=38, dis=2):
h, w = origin.shape[0], origin.shape[1]
rh = region_1.shape[0]
for pos in position:
center, radius = pos["center"], pos["radius"]
radius = 40
x = int(w * box[0] + center[0])
y = int(h * box[2] + rh - dis + 2 * gap + center[1])
cv2.circle(origin, (x, y), radius, (0, 0, 255), 3)
cv2.namedWindow("origin with diff", cv2.WINDOW_NORMAL)
cv2.imshow("origin with diff", origin)
cv2.waitKey(0)
# 自动点击
def auto_click(origin, region_1, box, position, gap=38, dis=2):
h, w = origin.shape[0], origin.shape[1]
rh = region_1.shape[0]
for pos in position:
center, radius = pos["center"], pos["radius"]
x = int(w * box[0] + center[0])
y = int(h * box[2] + rh - dis + 2 * gap + center[1])
os.system("adb.exe shell input tap %d %d" % (x, y))
logging.info("tap:(%d,%d)" % (x, y))
time.sleep(0.05)
# 主函数入口
def main(argv):
# 参数列表，程序运行需要提供的参数
# box = None # 裁剪原始图像的参数，分别为宽和高的比例倍
# gap = None # 图像中间间隔的一半大小
# dis = None # 图像移位，微调系数
# num = None # 显示差异的数量
# filter_sz = None # 滤波核大小
# auto_clicked=True
# 仅有一个参数，则使用默认参数
if len(argv) == 1:
box = (0.20, 0.93, 0.05, 0.95)
gap = 38
dis = 2
num = 5
filter_sz = 13
auto_clicked = "True"
else: # 多个参数时需要进行参数解析,参数使用等号分割
try:
# 设置参数
para_pairs = {}
paras = argv[1:] # 参数
for para in paras:
para_pairs[para.split("=")[0]] = para.split("=")[1]
# 参数配对
if "gap" in para_pairs.keys():
gap = int(para_pairs["gap"])
else:
gap = 38
if "box" in para_pairs.keys():
box = tuple([float(i) for i in para_pairs["box"][1:-1].split(",")])
else:
box = (0.20, 0.93, 0.05, 0.95)
if "dis" in para_pairs.keys():
dis = int(para_pairs["dis"])
else:
dis = 2
if "num" in para_pairs.keys():
num = int(para_pairs["num"])
else:
num = 5
if "filter_sz" in para_pairs.keys():
filter_sz = int(para_pairs["filter_sz"])
else:
filter_sz = 13
if "auto_clicked" in para_pairs.keys():
auto_clicked = para_pairs["auto_clicked"]
else:
auto_clicked = "True"
except IOError:
logging.info("参数出错，请重新输入！")
return
st = time.time()
try:
os.system("adb.exe exec-out screencap -p >screenshot.png")
convert_img("screenshot.png")
except IOError:
logging.info("从手机获取图片出错，请检查adb工具是否安装及手机是否正常连接！")
return
logging.info(">>>从手机截图用时：%0.2f 秒
" % (time.time() - st))
st = time.time()
try:
origin = cv2.imread("screenshot.png") # 原始图像
region_1, region_2, region = crop_image(origin, box=box, gap=gap, dis=dis)
diff_img = diff(region_1, region_2)
dis_img = dispose_region(diff_img)
position = contour_pos(dis_img, num=num, filter_size=filter_sz)
while len(position) < num and filter_sz > 3:
filter_sz -= 1
position = contour_pos(dis_img, num=num, filter_size=filter_sz)
except IOError:
logging.info("处理图片出错！")
return
try:
if auto_clicked is "True":
threading.Thread(target=auto_click, args=(origin, region_1, box, position, gap, dis)).start()
except IOError:
logging.info(">>>尝试点击出错！")
logging.info(">>>处理图片用时：%0.2f 秒
" % (time.time() - st))
try:
dip_diff(origin, region, region_1, region_2, dis_img, position, box)
# draw_circle(origin, region_1, position, box, gap=gap, dis=dis)
except IOError:
logging.info("重组显示出错！")
return
if __name__ == "__main__":
if not DEBUG:
while True:
main(sys.argv)
else:
box = (0.19, 0.95, 0.05, 0.95)
gap = 38
dis = 2
num = 5
filter_sz = 13
origin = cv2.imread("c:/users/fanyu/desktop/adb/screenshot.png") # 原始图像
region_1, region_2, region = crop_image(origin, box=box, gap=gap, dis=dis)
cv2.namedWindow("", cv2.WINDOW_NORMAL)
cv2.imshow("", region_2)
diff_img = diff(region_1, region_2)
dis_img = dispose_region(diff_img)
cv2.namedWindow(" ", cv2.WINDOW_NORMAL)
cv2.imshow(" ", region_1)
cv2.imshow("", dis_img)
position = contour_pos(dis_img, num=num, filter_size=filter_sz)
dip_diff(origin, region, region_1, region_2, dis_img, position, box)
# draw_circle(origin, region_1, position, box, gap=gap, dis=dis)
另外，可到我的github下载完整版：
https://github.com/yooongchun/auto_find_difference
也可以到微信公众号查看完整的文章：yooongchun小屋