一、思路分析
首先拿到答题卡照片的时候,需要对照片进行一系列预处理操作,通过透视变换将图像摆正方便后续的操作。每一道题五个选项,有五道题,通过字典存放准确答案。没有依次对答题卡进行轮廓检测,这里采用的是正方形,宽高比是1:1,当然也可以是矩形,也可以通过指定其他的筛选进行进行过滤筛选。最后通过掩膜操作,因为用户所选择的答案都是被涂过的,也就是通过判断黑色和白色来进行区分是否是用户选择的答案。一行一行的存储,因为一道题是五个选项,每一行是一道题,这里采用从上到下从左到右分别依次存放1-5题的A-E选项。通过与标准答案字典所存放的索引进行对比,从而给出用户得分。
二、导包及其相关函数
#导入工具包
import numpy as np
import argparse
import imutils
import cv2
# 设置参数
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image", required=True,help="path to the input image")
args = vars(ap.parse_args())# 正确答案
ANSWER_KEY = {0: 1, 1: 4, 2: 0, 3: 3, 4: 1}#存放正确答案BEADB
def order_points(pts):# 一共4个坐标点rect = np.zeros((4, 2), dtype = "float32")# 按顺序找到对应坐标0123分别是 左上,右上,右下,左下# 计算左上,右下s = pts.sum(axis = 1)rect[0] = pts[np.argmin(s)]rect[2] = pts[np.argmax(s)]# 计算右上和左下diff = np.diff(pts, axis = 1)rect[1] = pts[np.argmin(diff)]rect[3] = pts[np.argmax(diff)]return rectdef four_point_transform(image, pts):#透视变换# 获取输入坐标点rect = order_points(pts)(tl, tr, br, bl) = rect#拿到答题卡四个顶点坐标# 计算输入的w和h值widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))#因为拿到的答题卡的不一定是正儿八经的矩形,需要对四条边进行计算长度,选出最长的作为最后矩形的长度widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))maxWidth = max(int(widthA), int(widthB))heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))maxHeight = max(int(heightA), int(heightB))# 变换后对应坐标位置dst = np.array([[0, 0],[maxWidth - 1, 0],[maxWidth - 1, maxHeight - 1],[0, maxHeight - 1]], dtype = "float32")#看个人需求而定,这里将图像左上角规定为(0,0)位置# 计算变换矩阵M = cv2.getPerspectiveTransform(rect, dst)warped = cv2.warpPerspective(image, M, (maxWidth, maxHeight))# 返回变换后结果return warped
def sort_contours(cnts, method="left-to-right"):#从上到下进行排序,因为题目就是一行一行的reverse = Falsei = 0if method == "right-to-left" or method == "bottom-to-top":reverse = Trueif method == "top-to-bottom" or method == "bottom-to-top":i = 1boundingBoxes = [cv2.boundingRect(c) for c in cnts](cnts, boundingBoxes) = zip(*sorted(zip(cnts, boundingBoxes), key=lambda b: b[1][i], reverse=reverse))#排完序之后,前五个是第一题的,之后每五个依次为下一题的return cnts, boundingBoxes
def cv_show(name,img):cv2.imshow(name, img)cv2.waitKey(0)cv2.destroyAllWindows()
三、对答题卡进行预处理及透视变换摆正
# 预处理
image = cv2.imread(args["image"])#读取图像
contours_img = image.copy()#为了不改动原始图像,copy一下图像,因为后续需要进行一系列轮廓检测
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)#灰度图
blurred = cv2.GaussianBlur(gray, (5, 5), 0)#高斯滤波,去除一些噪音点
cv_show('blurred',blurred)
edged = cv2.Canny(blurred, 75, 200)#Canny边缘检测
cv_show('edged',edged)
# 轮廓检测
cnts = cv2.findContours(edged.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]#轮廓检测完之后会得到三个返回值,这里的[1]存放的是轮廓信息
cv2.drawContours(contours_img,cnts,-1,(0,0,255),3) #将图像通过透视变换进行摆正,绘制出答题卡的大致轮廓,拿到的答题卡图像也不一定是正儿八经的矩形
cv_show('contours_img',contours_img)
docCnt = None
# 确保检测到了
if len(cnts) > 0:#因为可能会检测到其他干扰影响,但是答题卡的轮廓是最大的# 根据轮廓大小进行排序cnts = sorted(cnts, key=cv2.contourArea, reverse=True)#把检测到的所有轮廓按面积进行排序# 遍历每一个轮廓for c in cnts:# 近似peri = cv2.arcLength(c, True)#计算一下轮廓周长approx = cv2.approxPolyDP(c, 0.02 * peri, True)#对轮廓进行近似# 准备做透视变换if len(approx) == 4:#多边形顶点有四个也就是矩形,这个就是我们的答题卡轮廓docCnt = approxbreak
# 执行透视变换
warped = four_point_transform(gray, docCnt.reshape(4, 2))
cv_show('warped',warped)
四、对每一道题均进行轮廓检测,遍历筛选
# 自适应阈值处理
thresh = cv2.threshold(warped, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
cv_show('thresh',thresh)
thresh_Contours = thresh.copy()
# 找到每一道题轮廓
cnts = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]#这里再次进行轮廓检测,之所以不用霍夫圆检测是因为有可能答题卡会被全部涂满甚至越界
cv2.drawContours(thresh_Contours,cnts,-1,(0,0,255),3)
cv_show('thresh_Contours',thresh_Contours)
questionCnts = []
# 遍历,对所有的轮廓进行筛选
for c in cnts:# 计算比例和大小(x, y, w, h) = cv2.boundingRect(c)ar = w / float(h)#因为是圆形,这里是宽高比,外接矩形差不多宽高比是1:1# 根据实际情况指定标准if w >= 20 and h >= 20 and ar >= 0.9 and ar <= 1.1:questionCnts.append(c)
# 按照从上到下进行排序
questionCnts = sort_contours(questionCnts, method="top-to-bottom")[0]
correct = 0
五、对比答案,评分
# 每排有5个选项
for (q, i) in enumerate(np.arange(0, len(questionCnts), 5)):#因为每一题都有五个选项,q为第几行# 排序cnts = sort_contours(questionCnts[i:i + 5])[0]#第i题的五个结果bubbled = None# 遍历每一个结果for (j, c) in enumerate(cnts):#j为第i道题的第j个选项# 使用mask来判断结果mask = np.zeros(thresh.shape, dtype="uint8")cv2.drawContours(mask, [c], -1, 255, -1) #-1表示填充cv_show('mask',mask)# 通过计算非零点数量来算是否选择这个答案mask = cv2.bitwise_and(thresh, thresh, mask=mask)total = cv2.countNonZero(mask)#看下框出来的选项中非零的个数有多少个# 通过阈值判断if bubbled is None or total > bubbled[0]:bubbled = (total, j)# 对比正确答案color = (0, 0, 255)k = ANSWER_KEY[q]#第q道题的答案# 判断正确if k == bubbled[1]:color = (0, 255, 0)correct += 1# 绘图cv2.drawContours(warped, [cnts[k]], -1, color, 3)
score = (correct / 5.0) * 100
print("[INFO] score: {:.2f}%".format(score))
cv2.putText(warped, "{:.2f}%".format(score), (10, 30),cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 0, 255), 2)
cv2.imshow("Original", image)
cv2.imshow("Exam", warped)
cv2.waitKey(0)
六、Pycharm参数设定
设置参数指定图像路径
找到Edit Configurations
将image参数改成自己测试图像路径--image images\test11.png
,其中images\test11.png为答题卡路径
七、完整代码
#导入工具包
import numpy as np
import argparse
import imutils
import cv2# 设置参数
ap = argparse.ArgumentParser()
ap.add_argument("-i", "--image", required=True,help="path to the input image")
args = vars(ap.parse_args())# 正确答案
ANSWER_KEY = {0: 1, 1: 4, 2: 0, 3: 3, 4: 1}#存放正确答案BEADBdef order_points(pts):# 一共4个坐标点rect = np.zeros((4, 2), dtype = "float32")# 按顺序找到对应坐标0123分别是 左上,右上,右下,左下# 计算左上,右下s = pts.sum(axis = 1)rect[0] = pts[np.argmin(s)]rect[2] = pts[np.argmax(s)]# 计算右上和左下diff = np.diff(pts, axis = 1)rect[1] = pts[np.argmin(diff)]rect[3] = pts[np.argmax(diff)]return rectdef four_point_transform(image, pts):#透视变换# 获取输入坐标点rect = order_points(pts)(tl, tr, br, bl) = rect#拿到答题卡四个顶点坐标# 计算输入的w和h值widthA = np.sqrt(((br[0] - bl[0]) ** 2) + ((br[1] - bl[1]) ** 2))#因为拿到的答题卡的不一定是正儿八经的矩形,需要对四条边进行计算长度,选出最长的作为最后矩形的长度widthB = np.sqrt(((tr[0] - tl[0]) ** 2) + ((tr[1] - tl[1]) ** 2))maxWidth = max(int(widthA), int(widthB))heightA = np.sqrt(((tr[0] - br[0]) ** 2) + ((tr[1] - br[1]) ** 2))heightB = np.sqrt(((tl[0] - bl[0]) ** 2) + ((tl[1] - bl[1]) ** 2))maxHeight = max(int(heightA), int(heightB))# 变换后对应坐标位置dst = np.array([[0, 0],[maxWidth - 1, 0],[maxWidth - 1, maxHeight - 1],[0, maxHeight - 1]], dtype = "float32")#看个人需求而定,这里将图像左上角规定为(0,0)位置# 计算变换矩阵M = cv2.getPerspectiveTransform(rect, dst)warped = cv2.warpPerspective(image, M, (maxWidth, maxHeight))# 返回变换后结果return warped
def sort_contours(cnts, method="left-to-right"):#从上到下进行排序,因为题目就是一行一行的reverse = Falsei = 0if method == "right-to-left" or method == "bottom-to-top":reverse = Trueif method == "top-to-bottom" or method == "bottom-to-top":i = 1boundingBoxes = [cv2.boundingRect(c) for c in cnts](cnts, boundingBoxes) = zip(*sorted(zip(cnts, boundingBoxes), key=lambda b: b[1][i], reverse=reverse))#排完序之后,前五个是第一题的,之后每五个依次为下一题的return cnts, boundingBoxes
def cv_show(name,img):cv2.imshow(name, img)cv2.waitKey(0)cv2.destroyAllWindows() # 预处理
image = cv2.imread(args["image"])#读取图像
contours_img = image.copy()#为了不改动原始图像,copy一下图像,因为后续需要进行一系列轮廓检测
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)#灰度图
blurred = cv2.GaussianBlur(gray, (5, 5), 0)#高斯滤波,去除一些噪音点
cv_show('blurred',blurred)
edged = cv2.Canny(blurred, 75, 200)#Canny边缘检测
cv_show('edged',edged)# 轮廓检测
cnts = cv2.findContours(edged.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]#轮廓检测完之后会得到三个返回值,这里的[1]存放的是轮廓信息
cv2.drawContours(contours_img,cnts,-1,(0,0,255),3) #将图像通过透视变换进行摆正,绘制出答题卡的大致轮廓,拿到的答题卡图像也不一定是正儿八经的矩形
cv_show('contours_img',contours_img)
docCnt = None# 确保检测到了
if len(cnts) > 0:#因为可能会检测到其他干扰影响,但是答题卡的轮廓是最大的# 根据轮廓大小进行排序cnts = sorted(cnts, key=cv2.contourArea, reverse=True)#把检测到的所有轮廓按面积进行排序# 遍历每一个轮廓for c in cnts:# 近似peri = cv2.arcLength(c, True)#计算一下轮廓周长approx = cv2.approxPolyDP(c, 0.02 * peri, True)#对轮廓进行近似# 准备做透视变换if len(approx) == 4:#多边形顶点有四个也就是矩形,这个就是我们的答题卡轮廓docCnt = approxbreak# 执行透视变换
warped = four_point_transform(gray, docCnt.reshape(4, 2))
cv_show('warped',warped)# 自适应阈值处理
thresh = cv2.threshold(warped, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
cv_show('thresh',thresh)
thresh_Contours = thresh.copy()# 找到每一道题轮廓
cnts = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]#这里再次进行轮廓检测,之所以不用霍夫圆检测是因为有可能答题卡会被全部涂满甚至越界
cv2.drawContours(thresh_Contours,cnts,-1,(0,0,255),3)
cv_show('thresh_Contours',thresh_Contours)
questionCnts = []# 遍历,对所有的轮廓进行筛选
for c in cnts:# 计算比例和大小(x, y, w, h) = cv2.boundingRect(c)ar = w / float(h)#因为是圆形,这里是宽高比,外接矩形差不多宽高比是1:1# 根据实际情况指定标准if w >= 20 and h >= 20 and ar >= 0.9 and ar <= 1.1:questionCnts.append(c)# 按照从上到下进行排序
questionCnts = sort_contours(questionCnts, method="top-to-bottom")[0]
correct = 0# 每排有5个选项
for (q, i) in enumerate(np.arange(0, len(questionCnts), 5)):#因为每一题都有五个选项,q为第几行# 排序cnts = sort_contours(questionCnts[i:i + 5])[0]#第i题的五个结果bubbled = None# 遍历每一个结果for (j, c) in enumerate(cnts):#j为第i道题的第j个选项# 使用mask来判断结果mask = np.zeros(thresh.shape, dtype="uint8")cv2.drawContours(mask, [c], -1, 255, -1) #-1表示填充cv_show('mask',mask)# 通过计算非零点数量来算是否选择这个答案mask = cv2.bitwise_and(thresh, thresh, mask=mask)total = cv2.countNonZero(mask)#看下框出来的选项中非零的个数有多少个# 通过阈值判断if bubbled is None or total > bubbled[0]:bubbled = (total, j)# 对比正确答案color = (0, 0, 255)k = ANSWER_KEY[q]#第q道题的答案# 判断正确if k == bubbled[1]:color = (0, 255, 0)correct += 1# 绘图cv2.drawContours(warped, [cnts[k]], -1, color, 3)score = (correct / 5.0) * 100
print("[INFO] score: {:.2f}%".format(score))
cv2.putText(warped, "{:.2f}%".format(score), (10, 30),cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 0, 255), 2)
cv2.imshow("Original", image)
cv2.imshow("Exam", warped)
cv2.waitKey(0)
答题卡原题:
这里展示的太大,我就截取了其中一小部分进行展示
接下来就是依次对每道题进行遍历找到掩膜,一共25次,这里就不一一展示了
最后根据设定的字典里面的正确答案给出评分