OpenCV+Python识别机读卡

背景介绍

正常机读卡是通过读卡机读取识别结果的，目前OpenCV已经这么强大了，尝试着用OpenCV+Python来识别机读卡。要识别的机读卡长这样：

我们做以下操作：

1.识别答题卡中每题选中项结果。

不做以下操作：

1.不识别101-106题（这些题实际情况下经常用不到，如果要识别原理也一样）

实现思路

通过分析答题卡特征：

1.答题区域为整张图片最大轮廓，先找出答题区域。

2.答题区域分为6行，每行4组，第6行只有1组，我们暂不处理第6行，只处理前面5行。

3.给定每一行第一个选项中心点坐标，该行其余选项的中心点坐标可以推算出来。

4.通过找到每个选项中心点坐标，再加上选项宽高，就可以在答题区域绘出每个选项的范围。

5.通过计算每个选项范围图像里非0像素点个数：

   单选题非0像素点最少的既是答案。

   多选题结合阈值判断该选项是否选中。

6.输出完整答案。

实现步骤

1.图像预处理

将图片转灰度图、黑帽运算：移除干扰项、二值化突出轮廓、查找轮廓、找出答题区域。

import cv2
import numpy as np# 1.读取图片并缩放
orginImg = cv2.imread("1.jpg")
size = ((int)(650*1.8), (int)(930*1.8))  # 尽可能将图片弄大一点，下面好处理
img = cv2.resize(orginImg, size)#显示图像
def imshow(name,image):scale_percent = 50  # 缩放比例width = int(image.shape[1] * scale_percent / 100)height = int(image.shape[0] * scale_percent / 100)dim = (width, height)resized_image = cv2.resize(image, dim, interpolation=cv2.INTER_AREA)cv2.imshow(name, resized_image)imshow("1.orgin", img)# 2.转灰度图
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
imshow("2.gray", gray)# 3.黑帽运算：移除干扰项
cvblackhat = cv2.morphologyEx(gray, cv2.MORPH_BLACKHAT, cv2.getStructuringElement(cv2.MORPH_RECT, (15, 15)))
imshow("3.black", cvblackhat)# 4.二值化突出轮廓，自动阈值范围 cv2.THRESH_BINARY|cv2.THRESH_OTSU
thresh = cv2.threshold(cvblackhat, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
imshow("4.thresh", thresh)# 5.提取轮廓，并在图上标记轮廓
cnts, hierarchy = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
mark = img.copy()
cv2.drawContours(mark, cnts, -1, (0, 0, 255), 2)
imshow("5.contours", mark)# 6.提取我们感兴趣的部分（这里我们只需要答题部分） img[y:y+h, x:x+w]
roi = None
for (i, c) in enumerate(cnts):(x, y, w, h) = cv2.boundingRect(c)ar = w/float(h)if w > 500 and h > 500 and ar > 0.9 and ar < 1.1:roi = img[y:y+h, x:x+w]break
imshow("5.roi", roi)

运行结果

2.查找每个选项的中心点坐标

这里每行起始点坐标和每个选项宽高，都是写的固定值（手动量出来的，本来想通过图像提取轮廓来取每个选项中心坐标点的，可是由于机读卡填图不规范，可能会影响轮廓提取，不是很靠谱，暂时没想到更好的办法）

# 7.查找每个选项的中心点坐标
# 思路：
# 通过分析：
# 1.答题区域分为6行，每行4组，第6行只有1组，我们暂不处理第6行，只处理前面5行。
# 2.只要给定每一行第一个选项中心坐标，该行其余选项的中心坐标可以推算出来。
# 3.通过找到每个选项中心点坐标，再加上选项宽高，就可以在答题区域绘出每个选项的范围。
# 4.通过计算每个选项范围图像里非0像素点个数，结合阈值判断该选项是否选中。
# 5.结合题目个数，遍历每个选项，构造出最终答案。
item = [34, 20]  # 每个选项宽度（跟图形缩放有关系）
x_step = 44  # x方向行距（两个选项水平方向距离）
y_step = 28  # y方向行距（两个选项垂直方向距离）
blank = 92  # 每组间距(5个一组)水平方向距离
centers = []  # 每个选项的中心点坐标，用来框选选项
# 答题区域有5行多1组，这里只处理前面5行，最后一组暂不处理
startPonits = [(25, 44), (25, 216), (26, 392), (26, 566), (28, 744)]
for (i, p) in enumerate(startPonits):temp = []  # 暂存该组选项坐标start = list(p)  # 该行起始点坐标for g in range(0, 4, 1):  # 每行有4组if len(temp) > 0:startx = temp[len(temp)-1][0] + blank  # 最后一个选项的x坐标+每组间距else:startx = start[0]start[0] = startxfor i in range(start[0], start[0]+5*x_step, x_step):  # 水平5个选项for j in range(start[1], start[1]+4*y_step, y_step):  # 垂直4个选项temp.append((i, j))for (i, c) in enumerate(temp):centers.append(c)# 8.将选项绘制到答题区域
show = roi.copy()
for (i, (x, y)) in enumerate(centers):left = x-(int)(item[0]/2)top = y-(int)(item[1]/2)# 绘选项区域矩形cv2.rectangle(show, (left, top), (left +item[0], top + item[1]), (0, 0, 255), -1)# 绘制序号标签cv2.putText(show, str(i+1), (left, top+10),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)
imshow("5.show", show)

运行结果

3.遍历每个选项,计算非0像素点个数

# 9.截取每个答题选项，并计算非0像素点个数
map = roi.copy()
map = cv2.cvtColor(map, cv2.COLOR_BGR2GRAY)
map = cv2.threshold(map, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
points = []
for (i, (x, y)) in enumerate(centers):left = x-(int)(item[0]/2)top = y-(int)(item[1]/2)# 截取每个选项小图item_img = map[top:top+item[1], left:left+item[0]]count = cv2.countNonZero(item_img)points.append(count)cv2.imwrite("item/"+str(i+1)+"-"+str(count)+".jpg", item_img)

这里为了方便观看，将每个选项截取另存为成图像了，以索“引号+非0像素点个数”命名。实际过程中可以不另存为图像。

大概像这样：

4.整理答案

# 10.整理答案
answer = []  # 二维数组：保存每个题ABCD4个选项值
group = []  # 将点分组，每4个1组，对应每题的4个选项
for i in range(0, len(points), 1):group.append(points[i])if (i+1) % 4 == 0:answer.append(group)group = []def printItem(i, optoins):question = {0: "A", 1: "B", 2: "C", 3: "D"}index = 0if i < 80:# 单选题（非0像素点最少的既是答案）an = min(optoins)index = optoins.index(an)print("第"+str(i+1)+"题"+question.get(index))else:# 多选题（根据阈值来判断是否选中）ans = ""for (j, p) in enumerate(options):if p < 400:index = optoins.index(p)ans += question.get(index)print("第"+str(i+1)+"题"+ans)# 打印题目答案
for (i, options) in enumerate(answer):printItem(i, options)

输出结果：

完整代码

import cv2
import numpy as np# 1.读取图片并缩放
orginImg = cv2.imread("1.jpg")
size = ((int)(650*1.8), (int)(930*1.8))  # 尽可能将图片弄大一点，下面好处理
img = cv2.resize(orginImg, size)# 显示图像
def imshow(name, image):scale_percent = 50  # 缩放比例width = int(image.shape[1] * scale_percent / 100)height = int(image.shape[0] * scale_percent / 100)dim = (width, height)resized_image = cv2.resize(image, dim, interpolation=cv2.INTER_AREA)cv2.imshow(name, resized_image)imshow("1.orgin", img)# 2.转灰度图
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
imshow("2.gray", gray)# 3.黑帽运算：移除干扰项
cvblackhat = cv2.morphologyEx(gray, cv2.MORPH_BLACKHAT, cv2.getStructuringElement(cv2.MORPH_RECT, (15, 15)))
imshow("3.black", cvblackhat)# 4.二值化突出轮廓，自动阈值范围 cv2.THRESH_BINARY|cv2.THRESH_OTSU
thresh = cv2.threshold(cvblackhat, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
imshow("4.thresh", thresh)# 5.提取轮廓，并在图上标记轮廓
cnts, hierarchy = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
mark = img.copy()
cv2.drawContours(mark, cnts, -1, (0, 0, 255), 2)
imshow("5.contours", mark)# 6.提取我们感兴趣的部分（这里我们只需要答题部分） img[y:y+h, x:x+w]
roi = None
for (i, c) in enumerate(cnts):(x, y, w, h) = cv2.boundingRect(c)ar = w/float(h)if w > 500 and h > 500 and ar > 0.9 and ar < 1.1:roi = img[y:y+h, x:x+w]break
imshow("5.roi", roi)# 7.查找每个选项的中心点坐标
# 思路：
# 通过分析：
# 1.答题区域分为6行，每行4组，第6行只有1组，我们暂不处理第6行，只处理前面5行。
# 2.只要给定每一行第一个选项中心坐标，该行其余选项的中心坐标可以推算出来。
# 3.通过找到每个选项中心点坐标，再加上选项宽高，就可以在答题区域绘出每个选项的范围。
# 4.通过计算每个选项范围图像里非0像素点个数，结合阈值判断该选项是否选中。
# 5.结合题目个数，遍历每个选项，构造出最终答案。
item = [34, 20]  # 每个选项宽度（跟图形缩放有关系）
x_step = 44  # x方向行距（两个选项水平方向距离）
y_step = 28  # y方向行距（两个选项垂直方向距离）
blank = 92  # 每组间距(5个一组)水平方向距离
centers = []  # 每个选项的中心点坐标，用来框选选项
# 答题区域有5行多1组，这里只处理前面5行，最后一组暂不处理
startPonits = [(25, 44), (25, 216), (26, 392), (26, 566), (28, 744)]
for (i, p) in enumerate(startPonits):temp = []  # 暂存该组选项坐标start = list(p)  # 该行起始点坐标for g in range(0, 4, 1):  # 每行有4组if len(temp) > 0:startx = temp[len(temp)-1][0] + blank  # 最后一个选项的x坐标+每组间距else:startx = start[0]start[0] = startxfor i in range(start[0], start[0]+5*x_step, x_step):  # 水平5个选项for j in range(start[1], start[1]+4*y_step, y_step):  # 垂直4个选项temp.append((i, j))for (i, c) in enumerate(temp):centers.append(c)# 8.将选项绘制到答题区域
show = roi.copy()
for (i, (x, y)) in enumerate(centers):left = x-(int)(item[0]/2)top = y-(int)(item[1]/2)# 绘选项区域矩形cv2.rectangle(show, (left, top), (left +item[0], top + item[1]), (0, 0, 255), -1)# 绘制序号标签cv2.putText(show, str(i+1), (left, top+10),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)
cv2.imshow("5.show", show)# 9.截取每个答题选项，并计算非0像素点个数
map = roi.copy()
map = cv2.cvtColor(map, cv2.COLOR_BGR2GRAY)
map = cv2.threshold(map, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
points = []
for (i, (x, y)) in enumerate(centers):left = x-(int)(item[0]/2)top = y-(int)(item[1]/2)# 截取每个选项小图item_img = map[top:top+item[1], left:left+item[0]]count = cv2.countNonZero(item_img)points.append(count)cv2.imwrite("item/"+str(i+1)+"-"+str(count)+".jpg", item_img)# 10.整理答案
answer = []  # 二维数组：保存每个题ABCD4个选项值
group = []  # 将点分组，每4个1组，对应每题的4个选项
for i in range(0, len(points), 1):group.append(points[i])if (i+1) % 4 == 0:answer.append(group)group = []def printItem(i, optoins):question = {0: "A", 1: "B", 2: "C", 3: "D"}index = 0if i < 80:# 单选题（非0像素点最少的既是答案）an = min(optoins)index = optoins.index(an)print("第"+str(i+1)+"题"+question.get(index))else:# 多选题（根据阈值来判断是否选中）ans = ""for (j, p) in enumerate(options):if p < 400:index = optoins.index(p)ans += question.get(index)print("第"+str(i+1)+"题"+ans)# 打印题目答案
for (i, options) in enumerate(answer):printItem(i, options)cv2.waitKey(0)
cv2.destroyAllWindows()

2024-8-26 补充：识别准考证区域

在轮廓检测时，将准考证区域也选出来。然后用同样的方法，将准考证号每个选项标记出来，根据非0像素点来识别。

完整代码（包含与标准答案比对出成绩）

import cv2
import numpy as np# 1.读取图片并缩放
orginImg = cv2.imread("22.jpg")
size = ((int)(650*1.8), (int)(930*1.8))  # 尽可能将图片弄大一点，下面好处理
img = cv2.resize(orginImg, size)# 显示图像def imshow(name, image):scale_percent = 50  # 缩放比例width = int(image.shape[1] * scale_percent / 100)height = int(image.shape[0] * scale_percent / 100)dim = (width, height)resized_image = cv2.resize(image, dim, interpolation=cv2.INTER_AREA)cv2.imshow(name, resized_image)# 图像预处理：灰度化、二值化、提取轮廓、提取答题区域def prepare(img):# 1.原图imshow("1.orgin", img)# 2.转灰度图gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)imshow("2.gray", gray)# 3.黑帽运算：移除干扰项cvblackhat = cv2.morphologyEx(gray, cv2.MORPH_BLACKHAT, cv2.getStructuringElement(cv2.MORPH_RECT, (15, 15)))imshow("3.black", cvblackhat)# 4.二值化突出轮廓，自动阈值范围 cv2.THRESH_BINARY|cv2.THRESH_OTSUthresh = cv2.threshold(cvblackhat, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]imshow("4.thresh", thresh)# 5.提取轮廓，并在图上标记轮廓cnts, hierarchy = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)mark = img.copy()cv2.drawContours(mark, cnts, -1, (0, 0, 255), 2)imshow("5.contours", mark)return cnts# 提取我们感兴趣的部分（这里我们只需要准考证部分和答题部分） img[y:y+h, x:x+w]
cnts = prepare(img)
answer_area = None  # 答题区域
zkzh_area = None    # 准考证号区域
for (i, c) in enumerate(cnts):(x, y, w, h) = cv2.boundingRect(c)ar = w/float(h)if w > 500 and h > 500 and ar > 0.9 and ar < 1.1:if answer_area is None:answer_area = img[y:y+h, x:x+w]elif w > 200 and h > 100 and ar > 1.2 and ar < 1.4:if zkzh_area is None:zkzh_area = img[y:y+h, x:x+w]if answer_area is not None and zkzh_area is not None:breakimshow("5.answer_area", answer_area)
cv2.imshow("6.zkzh_area", zkzh_area)# 解析准考证号def get_zkzh(zkzh_area):zkzh_item = [34, 20]  # 每个选项宽度（跟图形缩放有关系）zkzh_x_step = 45  # x方向行距（两个选项水平方向距离）zkzh_y_step = 30  # y方向行距（两个选项垂直方向距离）zkzh_centers = []  # 每个选项的中心点坐标，用来框选选项zkzh_start_ponits = (28, 112)for i in range(0, 11, 1):  # 11位准考证号startx = zkzh_start_ponits[0]+i*zkzh_x_stepfor j in range(0, 10, 1):  # 每位准考证号有10位数starty = zkzh_start_ponits[1]+j*zkzh_y_stepzkzh_centers.append((startx, starty))# 将选项绘制到准考证号zkzh_show = zkzh_area.copy()for (i, (x, y)) in enumerate(zkzh_centers):left = x-(int)(zkzh_item[0]/2)top = y-(int)(zkzh_item[1]/2)# 绘选项区域矩形cv2.rectangle(zkzh_show, (left, top), (left +zkzh_item[0], top + zkzh_item[1]), (0, 0, 255), -1)# 绘制序号标签cv2.putText(zkzh_show, str(i+1), (left, top+10),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)cv2.imshow("7.zkzh_show", zkzh_show)# 截取每个准考证号项，并计算非0像素点个数zkzh_map = zkzh_area.copy()zkzh_map = cv2.cvtColor(zkzh_map, cv2.COLOR_BGR2GRAY)zkzh_map = cv2.threshold(zkzh_map, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]zkzh_points = []for (i, (x, y)) in enumerate(zkzh_centers):left = x-(int)(zkzh_item[0]/2)top = y-(int)(zkzh_item[1]/2)# 截取每个选项小图item_img = zkzh_map[top:top+zkzh_item[1], left:left+zkzh_item[0]]count = cv2.countNonZero(item_img)zkzh_points.append(count)# 10.整理准考证号zkzh_nums = []  # 二维数组：保存每个题ABCD4个选项值zkzh_group = []  # 将点分组，每4个1组，对应每题的4个选项for i in range(0, len(zkzh_points), 1):zkzh_group.append(zkzh_points[i])if (i+1) % 10 == 0:zkzh_nums.append(zkzh_group)zkzh_group = []zkzh_num = ""for (i, zkzh_area) in enumerate(zkzh_nums):an = min(zkzh_area)index = zkzh_area.index(an)zkzh_num += str(index)return zkzh_numprint("准考证号：", get_zkzh(zkzh_area))# ============================解析答题区域=================================#
# 7.查找每个选项的中心点坐标
# 思路：
# 通过分析：
# 1.答题区域分为6行，每行4组，第6行只有1组，我们暂不处理第6行，只处理前面5行。
# 2.只要给定每一行第一个选项中心坐标，该行其余选项的中心坐标可以推算出来。
# 3.通过找到每个选项中心点坐标，再加上选项宽高，就可以在答题区域绘出每个选项的范围。
# 4.通过计算每个选项范围图像里非0像素点个数，结合阈值判断该选项是否选中。
# 5.结合题目个数，遍历每个选项，构造出最终答案。
answer_item = [34, 20]  # 每个选项宽度（跟图形缩放有关系）
answer_x_step = 44  # x方向行距（两个选项水平方向距离）
answer_y_step = 28  # y方向行距（两个选项垂直方向距离）
answer_blank = 92  # 每组间距(5个一组)水平方向距离
answer_centers = []  # 每个选项的中心点坐标，用来框选选项
# 答题区域有5行多1组，这里只处理前面5行，最后一组暂不处理
answer_start_ponits = [(25, 44), (25, 216), (26, 392), (26, 566), (28, 744)]
for (i, p) in enumerate(answer_start_ponits):temp = []  # 暂存该组选项坐标start = list(p)  # 该行起始点坐标for g in range(0, 4, 1):  # 每行有4组if len(temp) > 0:startx = temp[len(temp)-1][0] + answer_blank  # 最后一个选项的x坐标+每组间距else:startx = start[0]start[0] = startxfor i in range(start[0], start[0]+5*answer_x_step, answer_x_step):  # 水平5个选项for j in range(start[1], start[1]+4*answer_y_step, answer_y_step):  # 垂直4个选项temp.append((i, j))for (i, c) in enumerate(temp):answer_centers.append(c)# 8.将选项绘制到答题区域
answer_show = answer_area.copy()
for (i, (x, y)) in enumerate(answer_centers):left = x-(int)(answer_item[0]/2)top = y-(int)(answer_item[1]/2)# 绘选项区域矩形cv2.rectangle(answer_show, (left, top), (left +answer_item[0], top + answer_item[1]), (0, 0, 255), -1)# 绘制序号标签cv2.putText(answer_show, str(i+1), (left, top+10),cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)
cv2.imshow("7.answer_show", answer_show)# 9.截取每个答题选项，并计算非0像素点个数
answer_map = answer_area.copy()
answer_map = cv2.cvtColor(answer_map, cv2.COLOR_BGR2GRAY)
answer_map = cv2.threshold(answer_map, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
answer_points = []
for (i, (x, y)) in enumerate(answer_centers):left = x-(int)(answer_item[0]/2)top = y-(int)(answer_item[1]/2)# 截取每个选项小图item_img = answer_map[top:top+answer_item[1], left:left+answer_item[0]]count = cv2.countNonZero(item_img)answer_points.append(count)# cv2.imwrite("item/"+str(i+1)+"-"+str(count)+".jpg", item_img)# 9.截取每个答题选项，并计算非0像素点个数
answer_map = answer_area.copy()
answer_map = cv2.cvtColor(answer_map, cv2.COLOR_BGR2GRAY)
answer_map = cv2.threshold(answer_map, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
answer_points = []
for (i, (x, y)) in enumerate(answer_centers):left = x-(int)(answer_item[0]/2)top = y-(int)(answer_item[1]/2)# 截取每个选项小图item_img = answer_map[top:top+answer_item[1], left:left+answer_item[0]]count = cv2.countNonZero(item_img)answer_points.append(count)# 10.整理答案
answer = []  # 二维数组：保存每个题ABCD4个选项值
group = []  # 将点分组，每4个1组，对应每题的4个选项
for i in range(0, len(answer_points), 1):group.append(answer_points[i])if (i+1) % 4 == 0:answer.append(group)group = []# 最终结果
score_dic = {}def printItem(i, optoins):question = {0: "A", 1: "B", 2: "C", 3: "D"}index = 0if i < 80:# 单选题（非0像素点最少的既是答案）an = min(optoins)index = optoins.index(an)score_dic[i+1] = question.get(index)# print("第"+str(i+1)+"题"+question.get(index))else:# 多选题（根据阈值来判断是否选中）ans = ""for (j, p) in enumerate(options):if p < 400:index = optoins.index(p)ans += question.get(index)score_dic[i+1] = ans# print("第"+str(i+1)+"题"+ans)# 打印题目答案
for (i, options) in enumerate(answer):printItem(i, options)print("填涂结果：")
print(score_dic)# 与标准答案比较，得到最终得分
score = 0
answers = {1: "B", 2: "B", 3: "A", 4: "B", 5: "B", 6: "A", 7: "B", 8: "B", 9: "A", 10: "B", 11: "B", 12: "A", 13: "A", 14: "B", 15: "B", 16: "A", 17: "A", 18: "B", 19: "B", 20: "A", 21: "C", 22: "C", 23: "B", 24: "D", 25: "C", 26: "D", 27: "A", 28: "D", 29: "C", 30: "B", 31: "C", 32: "D", 33: "B", 34: "A", 35: "D", 36: "C", 37: "B", 38: "B", 39: "D", 40: "D", 41: "B",42: "C", 43: "B", 44: "D", 45: "B", 46: "A", 47: "D", 48: "D", 49: "C", 50: "D", 51: "B", 52: "D", 53: "A", 54: "A", 55: "D", 56: "D", 57: "C", 58: "D", 59: "B", 60: "D", 61: "A", 62: "B", 63: "D", 64: "B", 65: "C", 66: "D", 67: "B", 68: "C", 69: "D", 70: "A", 71: "ABCD", 72: "ABCD", 73: "AC", 74: "BCD", 75: "BC", 76: "BD", 77: "ABCD", 78: "ABCD", 79: "ACD", 80: "ABD"}
print("标准答案：")
print(answers)
for (i, a) in enumerate(score_dic):s = 0  # 每题得分if i > 0 and i < 20:s = 1.25  # 1-20题每题1.25分elif i > 20 and i < 70:s = 1     # 21-70题每题1分elif i > 70 and i < 80:s = 2.5   # 71-80题每题2.5分if answers.get(i+1) == score_dic.get(i+1):score += sprint("最终成绩：")
print(score)cv2.waitKey(0)
cv2.destroyAllWindows()

运行效果