# -*- coding:utf-8 -*-
from imutils.perspective import four_point_transform
from imutils import contours
import numpy as np
import cv2 as cv

# 加载原图,可在项目imgs/example02目录下找到
img = cv.imread("test01.jpg")

# cv.resizeWindow("enhanced", 240, 280);
# 打印原图
cv.imshow("orgin", img)

# 灰度化
gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)

# 打印灰度图
cv.imshow("gray", gray)

# 高斯滤波,清除一些杂点
blur = cv.GaussianBlur(gray, (3, 3), 0)

# 自适应二值化算法
thresh2 = cv.adaptiveThreshold(blur, 255, cv.ADAPTIVE_THRESH_GAUSSIAN_C, cv.THRESH_BINARY_INV, 131, 4)

# 打印二值化后的图
cv.imshow("thresh2", thresh2)

# 寻找轮廓
image, cts, hierarchy = cv.findContours(thresh2, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)

# 打印找到的轮廓
print("轮廓数:", len(cts))

# 对拷贝的原图进行轮廓标记
contour_flagged = cv.drawContours(img.copy(), cts, -1, (0, 0, 255), 3)
# 打印轮廓图
cv.imshow("contours_flagged", contour_flagged)
# 按像素面积降序排序
list = sorted(cts, key=cv.contourArea, reverse=True)

# 遍历轮廓
for ct in list:
 # 周长,第1个参数是轮廓,第二个参数代表是否是闭环的图形
 peri = 0.01 * cv.arcLength(ct, True)
 # 获取多边形的所有定点,如果是四个定点,就代表是矩形
 approx = cv.approxPolyDP(ct, peri, True)
 # 只考虑矩形
 if len(approx) == 4:

 # 从原图中提取所需的矫正图片
 ox = four_point_transform(img, approx.reshape(4, 2))
 # 从原图中提取所需的矫正图片
 tx = four_point_transform(gray, approx.reshape(4, 2))

 # 打印矫正后的灰度图
 cv.imshow("tx", tx)

 # 对矫正图进行高斯模糊
 blur = cv.GaussianBlur(tx, (3, 3), 0)

 # 对矫正图做自适应二值化
 thresh2 = cv.adaptiveThreshold(blur, 255, cv.ADAPTIVE_THRESH_GAUSSIAN_C, cv.THRESH_BINARY_INV, 131, 4)

 # 打印矫正后的二值化图
 cv.imshow("tx_thresh2", thresh2)

 # 获取轮廓
 r_image, r_cts, r_hierarchy = cv.findContours(thresh2, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)

 # 打印得到轮廓数量
 print("第二层轮廓数:", len(r_cts))

 # 用于存储答案的python list变量
 question_list = []
 for r_ct in r_cts:
 # 转为矩形,分别获取 x,y坐标,及矩形的宽和高
 x, y, w, h = cv.boundingRect(r_ct)

 # 过滤掉不符合答案坐标和长宽的选项
 if x > 2 and y > 2 and w > 20 and h > 20:
 # cv.drawContours(ox, r_ct, -1, (0, 0, 255), 1)
 question_list.append(r_ct)

 print("答案总数:", len(question_list))

 # 按坐标从上到下排序
 questionCnts = contours.sort_contours(question_list, method="top-to-bottom")[0]

 # 使用np函数,按5个元素,生成一个集合
 for (q, i) in enumerate(np.arange(0, len(questionCnts), 5)):

 # 每一个行5个答案,从左到右排序
 cnts = contours.sort_contours(questionCnts[i:i + 5])[0]

 # 存储一行题里面的每个答案
 ans_list = []
 for (j, cc) in enumerate(cnts):
 # 生成全黑画布
 mask = np.zeros(thresh2.shape, dtype="uint8")
 # 将每一个答案按轮廓写上去,并将填充颜色设置成白色
 tpp = cv.drawContours(mask, [cc], -1, 255, -1)
 # 两个图片做位运算
 mask = cv.bitwise_and(thresh2, thresh2, mask=mask)
 # 统计每个答案的像素
 total = cv.countNonZero(mask)

 # 添加到集合里面
 ans_list.append((total, j))

 # 按像素大小排序
 ans_list = sorted(ans_list, key=lambda x: x[0], reverse=True)

 max_ans_num = ans_list[0][1]
 max_ans_size = ans_list[0][0]
 print("答案序号:", max_ans_num, "列表:", ans_list)

 # 给选中答案,标记成红色
 cv.drawContours(ox, cnts[max_ans_num], -1, (0, 0, 255), 2)

 cv.imshow("answer_flagged", ox)

 # 最大的轮廓就是我们想要的,之后的就可以结束循环了
 break

# 阻塞等待窗体关闭
cv.waitKey(0)