基于opencv的答题卡识别

一、背景需求

传统的手动评分方法耗时且容易出错，自动化评分可以可以显著提高评分过程的速度和准确性、减少人工成本。
答题卡图片处理效果如下：

二、处理步骤

图片预处理

# 读图片
img = cv2.imread('./images/test_01.png')
cv_show('img', img)
# 变成黑白图片
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# 去掉一些噪点
blurred = cv2.GaussianBlur(gray, (5, 5), 0)
cv_show('blurred', blurred)

# 边缘检测
edged = cv2.Canny(blurred, 75, 200)
cv_show('edged', edged)

检测到答题卡轮廓

# 检测轮廓
cnts = cv2.findContours(edged.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]
# 画轮廓会修改被画轮廓的图. 
contours_img = img.copy()
cv2.drawContours(contours_img, cnts, -1, (0, 0, 255), 3)
cv_show('contous_img', contours_img)
# 确保我们拿到的轮廓是答题卡的轮廓.
if len(cnts) > 0:
    # 根据轮廓面积对轮廓进行排序.
    cnts = sorted(cnts, key=cv2.contourArea, reverse=True)
    
    # 遍历每一个轮廓
    for c in cnts:
        # 计算周长
        perimeter = cv2.arcLength(c, True)
        # 得到近似的轮廓
        approx = cv2.approxPolyDP(c, 0.02 * perimeter, True)
        # 近似完了之后, 应该只剩下4个角的坐标.
#         print(c)
#         print(approx)
        if len(approx) == 4:
            # 保存approx
            docCnt = approx
            # 找到答题卡近似轮廓, 直接推荐.
            break

透视变换

# 把透视变换功能封装成一个函数
def four_point_transform(image, pts):
    # 对输入的4个坐标排序
    rect = order_points(pts)
    (tl, tr, br, bl) = rect
    
    # 空间中两点的距离
    widthA = np.sqrt((br[0] - bl[0]) ** 2 + (br[1] - bl[1]) ** 2)
    widthB = np.sqrt((tr[0] - tl[0]) ** 2 + (tr[1] - tl[1]) ** 2)
    max_width = 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)
    max_height = max(int(heightA), int(heightB))
    
    # 构造变换之后的对应坐标位置.
    dst = np.array([
        [0, 0],
        [max_width - 1, 0],
        [max_width - 1, max_height - 1],
        [0, max_height - 1]], dtype='float32')
    
    # 计算变换矩阵
    M = cv2.getPerspectiveTransform(rect, dst)
    # 透视变换
    warped = cv2.warpPerspective(image, M, (max_width, max_height))
    return warped

透视变化后二值化：

thresh = cv2.threshold(warped, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1]
cv_show('thresh', thresh)

此时图片预处理好后如果需要 OCR 文本识别可借助 tesseract 工具识别文字

import pytesseract
from PIL import Image
# pytesseract要求的image不是opencv读进来的image, 而是pillow这个包, 即PIL,按照 pip install pillow
text = pytesseract.image_to_string(Image.open('./scan.jpg'))
print(text)

找每个圆圈的轮廓

# 找到每一个圆圈的轮廓
cnts = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)[1]
thresh_contours = thresh.copy()
cv2.drawContours(thresh_contours, cnts, -1, 255, 3)
cv_show('thresh_contours', thresh_contours)
plt.imshow(thresh_contours, cmap='gray')

# 遍历所有的轮廓, 找到特定宽高和特定比例的轮廓, 即圆圈的轮廓.
question_cnts = []
for c in cnts:
    # 找到轮廓的外接矩形
    (x, y, w, h) = cv2.boundingRect(c)
    # 计算宽高比
    ar = w / float(h)
    
    # 根据实际情况制定标准.
    if w >= 20 and h >= 20 and 0.9 <= ar <= 1.1:
        question_cnts.append(c)

轮廓排序

对轮廓按照y轴排序

# 轮廓排序功能封装成函数
def sort_contours(cnts, method='left-to-right'):
    reverse = False
    # 排序的时候, 取x轴数据, i=0, 取y轴数据i =1
    i = 0
    if method == 'right-to-left' or method == 'bottom-to-top':
        reverse = True
    # 按y轴坐标排序
    if method == 'top-to-bottom' or method == 'bottom-to-top':
        i = 1
    # 计算每个轮廓的外接矩形
    bounding_boxes = [cv2.boundingRect(c) for c in cnts]
    (cnts, bounding_boxes) = zip(*sorted(zip(cnts, bounding_boxes), key=lambda b: b[1][i], reverse=reverse))
    return cnts, bounding_boxes

# 按照从上到下的顺序对question_cnts排序.
question_cnts = sort_contours(question_cnts, method='top-to-bottom')[0]

判断是否答题正确

进一步按照x轴排序，构造圆圈轮廓的掩膜得出答案，最后和正确答案比较评判答题卡分数。

# 正确答案
ANSWER_KEY = {0:1, 1:4, 2:0, 3:3, 4:1}
correct = 0
for (q, i) in enumerate(np.arange(0, 25, 5)):
#     print(q, i)
    # 每次取出5个轮廓, 再按照x轴坐标从小到大排序
    cnts = sort_contours(question_cnts[i: i + 5])[0]
    
    bubbled = None
    # 遍历每一个结果
    for (j, c) in enumerate(cnts):
        # 使用掩膜, 即mask
        mask = np.zeros(thresh.shape, dtype='uint8')
        cv2.drawContours(mask, [c], -1, 255, -1) 
#         cv_show('mask', mask)

        # 先做与运算
        mask = cv2.bitwise_and(thresh, thresh, mask=mask)
#         cv_show('mask', mask)
        # 计算非零个数, 选择的选项, 非零个数比较多, 没选中的选项非零个数少一些
        total = cv2.countNonZero(mask)
        
        if bubbled is None or total > bubbled[0]:
            bubbled = (total, j)
            
    color = (0, 0, 255)
    k = ANSWER_KEY[q]
    # 判断是否做题正确
    if k == bubbled[1]:
        correct += 1
        color = (0, 255, 0)
    # 绘图
    cv2.drawContours(warped, [cnts[k]], -1, color, 3)
        
# 计算分数
print(correct)
score = (correct / 5.0) * 100
print(f'score: {score:.2f}%')
cv2.putText(warped, str(score) + '%', (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 0, 255), 2)
cv_show('result', warped)